201247577 六、發明說明: 【發明所屬之技術領域】 本發明係關於一種玻璃熔融爐内監視方法、玻璃熔融爐 操作方法、玻璃熔融爐内監視系統及玻璃物品之製造方法。 【先前技術】 於玻璃之製造步驟中,有將玻璃原料投入至玻璃熔融 爐,使上述原料於玻璃熔融爐内熔解之步驟。投入至玻璃 熔融爐内之原料為固體,於玻璃熔融爐内緩慢熔解。將投 入並於玻璃熔融爐内堆積之原料稱作批料堆(batch pile)。 批料堆係沿著熔解之原料即熔融玻璃之流向(即自玻璃炼 融爐之上游向下游),緩慢移動。又,由於批料堆係藉由熱 量而逐漸熔解,故而緩慢變小。由於批料堆之行為成為玻 璃:融爐之操作之域’故而自設置於玻璃熔融爐内之觀 察窗藉由目測而觀察玻璃熔融爐内之批料堆,或者進行草 繪。於觀察批料堆之情形時,相較炫融玻璃之表面(即液面) 為上方之部分成為觀察對象。 、又’提出多種不通過利用目測之觀察或草綠,而對玻璃 熔融爐内之觀察窗配置攝像機而監視批料堆之方法。 定例2於非專利文獻1中記載之技術中,對監視區域之決 ::中可進行直線檢測之霍夫―轉換…於非專 J文獻1中,記載有求出批料堆之佔有率。 寻 又’於專利文獻1中,記載有對批料 像時刻,對拙粗& & _ 仃攝像,於各攝 位置進行比:,之邊界線之位置或形狀或最下游 164119.doc 201247577 又,於專利文獻2令, 像圖像’根據上述圓像獲得位二::二订掃描而攝 對亮度特性線狀批料堆之存在H 據位置 又,於專利文獻3令,記載有與於玻璃 料相關之參數之測定或調節方法。 ㈣落解之原 ^作為自圖像中_特定物體之基本方法,有使像素 二值化之方法。二值化亦有多種方法,例如有特定與亮度 值相對應之像素之直方圖之谷而將像素分成2個等級之方 法。作為特定與亮度值相對應之像素之直方圖之谷之方 法已知有模式法或判別分析二值化法等。模式法係記載 於非專利文獻2、3中。㈣分析=值化法係記載於非專利 文獻3中。於判別分析二值化法中,於將直方圖分割成㈣ 等級時,以2個等級間之分離成為最佳之方式,決定閾值。 具體而言,決定與圖像中之背景區域與特定物體之區域相 關之等級内方差與等級間方差之方差比成&最大之閾值。 【先前技術文獻J ί專利文獻】 [專利文獻1] 曰本專利特開2009-161396號公報 [專利文獻2] 曰本專利特開昭59-44606號公報 [專利文獻3] 美國專利申請公開第2004/0079113號說明書 [非專利文獻1 164119.doc 201247577 [非專利文獻1]201247577 VI. Description of the Invention: [Technical Field] The present invention relates to a glass melting furnace monitoring method, a glass melting furnace operating method, a glass melting furnace monitoring system, and a glass article manufacturing method. [Prior Art] In the manufacturing step of the glass, there is a step of introducing the glass raw material into a glass melting furnace to melt the raw material in a glass melting furnace. The raw material charged into the glass melting furnace is solid and slowly melts in the glass melting furnace. The raw material which is poured and accumulated in the glass melting furnace is referred to as a batch pile. The batch reactor moves slowly along the flow direction of the molten material, i.e., the molten glass (i.e., upstream from the glass refining furnace). Further, since the batch reactor is gradually melted by the heat, it is gradually reduced. Since the behavior of the batch pile becomes the domain of operation of the glass furnace, the observation piles placed in the glass melting furnace are observed by visual inspection to observe the batch pile in the glass melting furnace or to perform sketching. When observing the batch pile, the upper part of the surface of the glazed glass (ie, the liquid surface) becomes the object of observation. Further, various methods have been proposed for monitoring the batch pile by arranging a camera on the observation window in the glass melting furnace without using visual observation or grass green. In the technique described in the non-patent document 1, in the technique described in the non-patent document 1, the Hough-conversion which can perform the line detection in the determination of the monitoring area is described in Non-Special J Document 1, and the occupancy rate of the batch pile is determined. In the patent document 1, it is described that for the batch image time, the image is taken for the image of the image, and the ratio is: the position or shape of the boundary line or the most downstream 164119.doc 201247577 Further, in Patent Document 2, the image image 'according to the above-described circular image, the bit 2:: two-scheduled scanning is used to capture the presence of the brightness data of the linear batch pile, and in the patent document 3, Determination or adjustment of parameters related to glass frit. (4) The original solution of the solution ^ As a basic method for the specific object from the image, there is a method of binarizing the pixel. There are also various methods for binarization, such as a method of dividing a pixel into two levels by having a valley of a histogram corresponding to a luminance value. A method of determining a valley of a histogram of a pixel corresponding to a luminance value is known as a pattern method or a discriminant analysis binarization method. The pattern method is described in Non-Patent Documents 2 and 3. (4) The analysis = value method is described in Non-Patent Document 3. In the discriminant analysis binarization method, when the histogram is divided into (four) levels, the threshold is determined by the separation between the two levels. Specifically, it is determined that the variance ratio between the intra-level variance and the inter-level variance associated with the background region in the image and the region of the specific object is the & maximum threshold. [Patent Document 1] Japanese Laid-Open Patent Publication No. 2009-161396 [Patent Document 2] Japanese Patent Application Laid-Open No. Hei 59-44606 (Patent Document 3) [Instruction No. 2004/0079113] [Non-Patent Document 1 164119.doc 201247577 [Non-Patent Document 1]
Emmanuel Obser, Stephane Lepert, Sylvie Lelandais, 「IMAGE PROCESSING FOR GLASS INDUSTRY」,「Proceedings 4th International Conference on Quality Control by Artificial Vision」: QCAV87, ISBN : 4-921073-01-5, 1998年 11 月 10 日 [非專利文獻2] 「閾(臨限)值處理」、[online]、平成16年6月14曰、SUGIMOTO Yoshitaka、[平成22年10月1曰檢索]、網際網路< URL : http://www.mm.media.kyoto-u.ac.jp/education/DIP/ WEBPAGE一SECTION/section7/node2.html> [非專利文獻3] 「2值化處理」、[online]、村上·泉田研究室HP製作委員會 (2001)、[平成22年10月4曰檢索]、網際網路< URL : http://ipr20.cs.ehime-u.ac.jp/column/gazo_syori/chapter4.html> 【發明内容】 [發明所欲解決之問題] 於對玻璃熔融爐之觀察窗配置攝像機而監視批料堆之情 形時,較佳為使得可持續玻璃熔融爐内之固定區域之觀 察,而正確地監視上述區域中之批料堆之狀態。 然而,有時於觀察窗之打掃等維護作業時攝像機之位置 及方向會發生偏移。如此一來,攝像機之攝像範圍亦發生 偏移。如此般,若攝像機之位置或方向發生變化,則批料 堆之狀態之經時變化之評價精度降低。 又,於熔解之原料之液面,因原料被加熱而產生泡。因此, 164119.doc 201247577 於攝像玻璃炫融爐内之批料堆之情形時,獲得以泡為背景之 批料堆之圖像。為了正確地監視轉堆之㈣,較佳為將圖 像内之泡與批料堆㈣’而自圖像内操取崎堆之部分。 又,較佳為於監視批料堆時,根據其監視結果確切把 握調節玻璃炫融爐之哪一運轉參數即可,對玻璃炫融爐進 行操作》 因此’本發明之目的在於提供i可良好地持續玻璃溶 融爐内之固^區域之觀察之玻璃炫融爐内監視方法及玻璃溶 融爐内監視m其目的在於提供—種實現如上所述之 良好之觀察狀態並且製造玻璃物品之玻璃物品之製造方法。 又’本發明之目的在於提供一種可使根據監視之批料堆 之狀態,調節玻璃熔融爐之哪一運轉參數即可明確化之玻 璃熔融爐操作方法。 [解決問題之技術手段] 本發明之玻璃熔融爐内監視方法之特徵在於其包含如下 步驟:圖像攝像步驟,其係圖像攝像機構對包含設置於玻 璃溶融爐内之基準圖案與在玻璃熔融爐内熔解之玻璃原料 之液面中之固定範圍之圖像進行攝像;區域擷取步驟,、其 係根據使用拍攝於圖像内之基準圖案之位置偏移而計算之 圖像攝像機構之姿勢,自所攝像之圖像内擷取符合固定範 圍之區m圖像作成步驟’其係'根據作為符合固定矿: 圍之區域而自複數之圖像中擷取之複數之擷取圖像,作: 成為堆積於玻璃熔融爐内之玻璃原料即批料堆之背景a 景圖像;背景除外圖像生成步驟,其係藉由針對每二: 164U9.doc -6- 201247577 進行自從所攝像之圖像中作為符合固定範圍之區域而擷取 之擷取圖像之像素之亮度值減去背景圖像中之對應像素之 亮度值之處理,生成自拍攝有批料堆及背景之狀態之擷取 圖像中將背景除外之背景除外圖像;及觀察資料算出步驟, 其係根據背景除外圖像,算出與批料堆相關之觀察資料。 亦可為如下之方法:於背景圖像作成步驟中,針對複數 之擷取圖像之每一對應像素或每一對應之區域計數符合 各亮度值之像素之數量,根據符合各亮度值之像素之計數 結果決定表示背景之亮度值,藉此作成背景圖像。 亦可為如下之方法:於背景除外圖像生成步驟中,針對 每一像素進行自從所攝像之圖像中作為符合固定範圍之區 域而擷取之擷取圖像之像素之亮度值減去背景圖像中之對 應像素之亮度值之處理,並將每一像素之減法結果二值 化’藉此生成背景除外圖像。 亦可為如下之方法:其包含如下步驟:背景圖像轉換步 驟,其係將背景圖像轉換成自肖液面冑向之上方觀察固定 範圍時之圖像;及㈣时轉換步驟,㈣將作為符合固 定範圍之區域而擷取之擷取圖像轉換成自與液面對向之上 方觀察該固定範圍時之圖像;於背景除外圖像生成步驟 中’進行自利㈣取圖像轉換㈣進行轉換後之摘取圖像 之亮度值減去利用背景圖像轉換步驟進行轉換後之背景圖 像中之對應像素之亮度值之處理;於觀察資料算出步驟 中,根據於背景除外圖像生成步驟中生成之背景除外圖像 算出觀察資料。 164119.doc 201247577Emmanuel Obser, Stephane Lepert, Sylvie Lelandais, "IMAGE PROCESSING FOR GLASS INDUSTRY", "Proceedings 4th International Conference on Quality Control by Artificial Vision": QCAV87, ISBN : 4-921073-01-5, November 10, 1998 [Non Patent Document 2] "Threshold (preceding) value processing", [online], June 14th, 2004, SUGIMOTO Yoshitaka, [Retrieval on October 1st, 2002], Internet < URL: http:/ /www.mm.media.kyoto-u.ac.jp/education/DIP/ WEBPAGE-SECTION/section7/node2.html> [Non-Patent Document 3] "Binary Processing", [online], Murakami Izumi Research HP Production Committee (2001), [Retrieved on October 4th, 2008], Internet < URL: http://ipr20.cs.ehime-u.ac.jp/column/gazo_syori/chapter4.html> SUMMARY OF THE INVENTION [Problems to be Solved by the Invention] When the camera is placed in the observation window of the glass melting furnace to monitor the batch pile, it is preferable to observe the fixed area in the glass melting furnace, and correctly Monitor the status of the batch heap in the above area. However, there are cases where the position and orientation of the camera shift during maintenance work such as cleaning of the observation window. As a result, the camera's imaging range is also offset. In this way, if the position or orientation of the camera changes, the evaluation accuracy of the state change of the batch pile is lowered. Further, in the liquid surface of the molten raw material, bubbles are generated due to heating of the raw material. Therefore, 164119.doc 201247577 obtained the image of the batch pile in the background of the bubble in the case of the batch pile in the camera glass melting furnace. In order to properly monitor the transfer (4), it is preferable to operate the bunker portion from the image within the image and the batch (4). Moreover, it is preferable to accurately monitor which operating parameter of the glass smelting furnace according to the monitoring result when monitoring the batch pile, and operate the glass smelting furnace. Therefore, the object of the present invention is to provide good i. The monitoring method in the glass smelting furnace and the monitoring in the glass melting furnace for the observation of the solid area in the glass melting furnace are to provide a glass article for realizing the good observation state as described above and manufacturing the glass article. Production method. Further, it is an object of the present invention to provide a glass melting furnace operation method which can clarify which operating parameter of a glass melting furnace can be adjusted according to the state of the monitored batch reactor. [Technical means for solving the problem] The glass melting furnace monitoring method of the present invention is characterized in that it comprises the following steps: an image capturing step of the image capturing means for melting the glass containing the reference pattern provided in the glass melting furnace An image of a fixed range in the liquid surface of the molten glass material in the furnace is imaged; a region capturing step is performed based on the positional deviation of the image capturing mechanism calculated using the reference pattern captured in the image , taking an image from the image captured in the image that meets the fixed range, the image forming step 'the system' is based on the image captured as a plurality of images taken from the complex image of the fixed mine: the surrounding area, The background image of the batch material piled up in the glass melting furnace is the background image of the batch pile; the background image generation step is performed by the camera for each of the two: 164U9.doc -6-201247577 The processing of subtracting the brightness value of the corresponding pixel in the background image from the brightness value of the pixel captured in the image as a region conforming to the fixed range, and generating the self-photographed batch and Except the background of the image capturing state in the background of the scene except for the image; and observations calculating step of excluding the background image based, calculates correlation of the batch reactor observations. The method may be as follows: in the background image forming step, counting the number of pixels corresponding to each brightness value for each corresponding pixel or each corresponding region of the plurality of captured images, according to the pixels that meet the respective brightness values The result of the counting determines the brightness value representing the background, thereby creating a background image. In the background exclusion image generation step, the luminance value of the pixel of the captured image captured from the captured image as the area corresponding to the fixed range is subtracted from the background for each pixel. The processing of the luminance values of the corresponding pixels in the image, and binarizing the subtraction result of each pixel', thereby generating a background exclusion image. The method may further include the following steps: a background image conversion step of converting the background image into an image when the fixed range is viewed from above the sloping surface; and (4) a time conversion step, (4) The captured image captured as a region conforming to the fixed range is converted into an image when the fixed range is viewed from above the liquid facing direction; in the background exclusion image generating step, 'self-interesting (four) is taken for image conversion (4) processing the brightness value of the extracted image after the conversion minus the brightness value of the corresponding pixel in the background image converted by the background image conversion step; in the observation data calculation step, according to the background exclusion image The background exclusion image generated in the generation step is used to calculate observation data. 164119.doc 201247577
亦可為如下之方法.甘& A , 、包3將背景除外圖像轉換成自與 液面對向之上方觀察固定銘 开 -^ _ 範圍時之圓像之背景除外圖像轉 換步驟,於觀察資料算出击 . /驟中,根據利用背景除外圖像 轉:步驟進行轉換後之背景除外圖像算出觀察資料。 像攝傻^包含如下步驟之方法:預處理步称,其係對在圖 像攝像步驟中所獲得之各 豕弄出表不圖像内之明暗之 2圓度:量,並選擇滿足對表示對比度之量預先規定之條 亦可為如下之方法:於預處理步驟中,算出圖像内之邊 緣數作為表示對比度之量,選擇滿足邊緣數為預先規定之 %值以上之條件之複數之圖像’根據所選擇之複數之圖 像’生成成為㈣符合㈣範圍之區域之對象之圖像。 又,本發明之玻璃熔融爐操作方法之特徵在於其包含如 下步驟‘影響度導出步驟,其係導出玻璃炫融爐之運轉參 數對在上述玻璃熔融爐内監視方法中之觀察資料算出步驟 中算出之觀察資料賦予之影響之程度;及熔融爐控制步 驟其係於觀察資料滿足特定條件之情形時,變更對該觀 察資料之影響之程度之絕對值成為預先規定之值以上之運 轉參數。 又’本發明之玻璃炫融爐内監視系統之特徵在於其包含 如下機構:圖像攝像機構,其對包含設置於玻璃熔融爐内 之基準圖案與在玻璃熔融爐内熔解之玻璃原料之液面中之 固定範圍之圖像進行攝像;圖像校準機構,其根據使用拍 攝;圖像内之基準圖案之位置偏移而計算之圖像攝像機構 1641l9.doc 201247577 =!:自所攝像之圖像内擁取符合固定範圍之區域;背 :圖像作成機構’其根據作為符合固定範圍之區域而自複 之圖像中揭取之複數之梅取圓像,作成成為堆積於玻璃 :融爐内之玻璃原料即批料堆之背景之背景圖像;差分運 機構’其藉由針對每—像素進行自從所攝像之圖像中作 為符合固定範圍之區域而揭取之榻取圖像之像素之亮度值 減去背景圖像中之對應像素之亮度值之處理,而生成自拍 攝有批料堆及#景之狀態之擷取圖像中將背景除外之背景 除外圖像’及觀察資料算出機構,其根據背景除外圖像, 算出與批料堆相關之觀察資料。 亦可為如下之構成:背景圖像作成機構針對複數之擷取 圖像之每一對應像素或每一對應之區域,計數符合各亮度 值之像素之數量,根據符合各亮度值之像素之計數結果, 決定表示背景之亮度值,藉此作成背景圓像。 亦可為如下之構成:差分運算機構針對每一像素進行自 從所攝像之圖像中作為符合固定範圍之區域而操取之掏取 圖像之像素之亮度值減去背景圖像中之對應像素之亮度值 之處理’並將每一像素之減法結果二值化,藉此生成背景 除外圖像》 亦可為如下之構成:圖像校準機構將背景圖像轉換成自 與液面對向之上方觀察固定範圍時之圖像,將作為符合固 定範圍之區域而擷取之擷取圖像轉換成自與液面對向之上 方觀察該固定範圍時之圖像,差分運算機構進行自利用圖 像校準機構進行轉換後之擷取圖像之亮度值減去利用圖像 164H9.doc 201247577 校準機構進行轉換後之背景圖像中之對應像素之亮度值之 處理,觀察資料算出機構根據藉由差分運算機構而生成之 背景除外圖像算出觀察資料。 亦可為如下之構成:圖像校準機構將藉由差分運算機構 而生成之背景除外圖像轉換成自與液面對向之上方觀察固 定範圍時之圖像,觀察資料算出機構根據利用圖像校準機 構進行轉換後之背景除外圖像算出觀察資料。 亦可為包含如下機構之構成:預處理機構,其對藉由圖像 攝像機構而獲得之各圖像,算出表示圖像内之明暗之對比度 之量,選擇滿足對表示對比度之量預先規定之條件之圖像。 亦可為如下之構成:預處理機構算出圖像内之邊緣數作 為表示對比度之量,選擇滿足邊緣數為預先規定之閾值以 上之條件之複數之圖像,根據所選擇之複數之圖像,生成 成為擷取符合固定範圍之區域之對象之圖像。 亦可為包含如下機構之構成:觀察資料解析機構,其導 出玻璃熔融爐之運轉參數對藉由觀察資料算出機構而算出 之觀察資料賦予之影響之程度。 亦可為包含如下機構之構成:熔融爐控制機構,其於觀 察1資料滿足特定條件之情形時,變更對該觀察資料之影響 之程度之絕對值成為預先規定之值以上之運轉參數。 又’本發明之玻璃物品之製造方法之特徵在於其包含如 下步驟:玻璃熔融步驟,其係於玻璃熔融爐内製造熔融玻 璃;澄清步驟,其係於澄清槽内去除熔融玻璃之泡; 成形步驟’其係成形將泡去除之熔融玻璃;及緩冷步驟’ I64119.doc • 10· 201247577 其係使所成形之熔融玻璃緩冷;並且包含如下步 攝?機構對包含設置於坡璃_内 /…軌融爐内炫解之玻璃原料之液面中之 固定範圍之圖像進行攝像; 拍攝於圖像内之基準圖宰之位步驟,其係根據使用 構之姿勢,自所攝像之計算之圖像攝像機 之圖像内擷取符合固定範圍之區域. 背景圖像作成步驟,其係根據作為符合固定範圍之區域而 中擷取之複數之摘取圓像,作成成為堆積於 景除㈣生成步景之背_;背 減去背景圖像中之對應像素之亮度值之處 理,而生成自拍攝有批料堆及背景之狀態: 背景除外之背景除外圖像;及觀察資料算出步驟,其2 據背景除外圖像’算出與批料堆相關之觀察資料。 [發明之效果] :據本發明之玻璃炫融爐内監視方法及 視系統,可持續玻璃溶融爐内之固定區域之觀察,而良ί 之批料堆之狀態…根據玻璃物 口口之製造方法’可實現如!_路 造玻璃H 料之良好之監視狀態並且製 二二發明之玻璃炫融爐操作方法,可使 爐之哪—運轉參數即可明辟。 164] 19.doc -11 - 201247577 以下’參照圖式對本發明之實施形態進行說明。 首先’對應用本發明之玻璃熔融爐内監視系統之玻璃熔 融爐之例進行說明。圖丨係表示如上所述之玻璃熔融爐之例 之俯視圖。玻璃熔融爐丨係於由底面、上游壁(上游側之 壁)7、側壁6、下游壁(下游側之壁)8及頂棚(省略圖示)包圍 之空間内,藉由熱量而使玻璃原料溶解。於上游壁7,設置 有投入原料之投入口 3a、3b,於下游壁8,設置有將熔解之 玻璃原料排出之排出口 4。又,於側壁6,分別設置有觀察 窗2與燃燒器5。於圖1中,表示設置有投入口 3a、3(}之情形, 但投入口之數量並不限定於2個。 自投入口 3a、3b投入固體狀態之玻璃原料。由於玻璃熔融 爐内由自燃燒器5噴出之火加熱,因此該原料緩慢熔解,熔 解之原料係緩慢移動至下游側而自排出口 4排出。於玻璃熔 融爐1内以固體狀態堆積之原料為批料堆丨〇。批料堆丨〇係隨 著時間經過而一面向下游側移動一面熔解。 本發明之玻璃熔融爐内監視系統包含攝像機11{1、11(),監 視玻璃熔融爐内之液面之固定區域9a、9b。於圖1中,例示 以爐内之液面中各攝像機之正面方向上之側壁間之區域由 2個固定區域9a、9b覆蓋之方式規定有2個固定區域9a、\之 情形。攝像機118係自上游側觀察而對右側之固定區域9a(以 下僅記作固定區域9a)進行攝像,攝像機llb係自上游側觀察 而對左側之固定區域9b(以下僅記作固定區域9b)進行攝 像。於本發明中,以玻璃熔融爐内監視系統包含2台攝像機 lla、nb之情形為例進行說明,但玻璃熔融爐内監視系統所 164119.doc 12 201247577 包含之攝像機之台數並不限定於2台。 再者固疋區域9a、9〇係與投入口 3a、3b附近分離而規定。 其原因在於:於將投人Wa、3b之較近之區域作為固定區域 而進行攝像之情形時’於攝像圖像内符合固定區域之部分 完全變為批料堆’成為背景之泡未拍攝之可能性較高,於 上述之情形時,無法算出與批料堆相關之資料。 [實施形態1] 圖2係表示本發明之第1實施形態之玻璃熔融爐内監視系 統之構成例的方塊圖。第1實施形態之玻璃熔融爐内監視系 統包含攝像機lla、攝像機llb及圖像處理裝置13。玻璃熔融 爐内監視系統分別對攝像機lla、⑴進行攝像所得之圖像進 行相同之處理。因此,以下,對攝像機lla進行說明適當 省略與攝像機llb相關之說明。 攝像機lla經由玻璃熔融爐之觀察窗2(參照圖丨),反覆攝 像液面之固定區域9a之圖像。該圖像為靜態圖像。同樣地, 攝像機llb亦經由玻璃熔融爐之觀察窗2(參照圖υ,反覆攝 像液面之固定區域9b之靜態圖像。攝像機i丨&、丨丨b之攝像間 隔預先規定即可。 再者’於攝像機lla之攝像範圍(視野之範圍),不僅收入 固定區域9a,亦收入固定區域\附近之液面或與攝像機Ua 對向之側壁《因此,於攝像機lla之攝像圖像中,亦拍攝有 固疋區域9a及其附近之液面或對向之側壁。關於攝像機1、 亦相同。 由攝像機11 a、11 b攝像之圖像係輸入至圖像處理裝置i 3。 164119.doc 5: •13· 201247577 理圖:=置13對由攝像機Ua攝像之圖像進行圖像處 =出與固定區域9a中之批料堆相關之多種資料(例如與 -置或動作相關之資料)。同 , )J樣也圖像處理裝置13對由攝 象機1U攝像之圖像進行圖像處理,算出與固定區域中之 =堆相關之多種資料。以下將根據攝像機lla、llb進行攝 所得之圖像而算出之批料堆之資料記作觀察資料。 圖像處理裝置13包含預處理機構19、圖像記憶機構12、 姿勢特定機構14、背景圖像作成機構15、圖像校準機構16、 差分運算機構17及觀察資料算出機構18。 預處理機構19根據攝像機lla進行攝像所得之圖像,生成 未拍攝有原料粉或火焰(自燃燒器5喷出之火)之狀態之圖 像。若!浮於玻璃溶融爐内之原#粉或火焰拍攝於圆像 中,則批料堆之圖像變得不清楚。預處理機構19使用攝像 機1U進行攝像所得之複數之圖像,生成未受原料粉或火焰 等干擾之影響而清楚地拍攝有批料堆之狀態之圖像。預處 理機構19對攝像機iib進行攝像所得之圖像亦進行相同之 處理。將如此般生成將原料粉或火焰之影響去除之圖像記 作預處理。又,有以下將預處理機構19根據藉由攝像機而 攝像之複數之圖像生成之圖像記作預處理圖像之情形。其 中’預處理圖像若除將原料粉或火焰之影響去除而使批料 堆更清楚之方面以外’則與各攝像機進行攝像所得之圖像 相同,亦有將預處理圖像僅記作攝像圖像之情形。即,有 與攝像機進行攝像所得之圖像本身同樣地稱作攝像圖像之 情形。預處理機構19分別將根據攝像機1丨3而獲得之預處理 164119.doc -14· 201247577 圖像及根據攝像機llb而獲得之預處理圖像記憶 憶機構12❶ 據破璃熔融爐’亦有完全不需要預處理或不需 要一部分之情形。例如,於火焰之影響較小或懸浮之原料 粉較少之玻璃熔融爐中,亦可不進行預處理。於上述之情 形時,圖像處縣置13將自錢像機Ub輸人之圖像直 接s己憶於圖像記憶機構12中即可。 圖像記憶機構⑽記憶圖像之記憶裝置。於如上所述, 預處理機構19對自各攝像機lu、^輸入之圖像進行預處理 之情形時,記憶藉由上述預處理而獲得之預處理圖像。又, ^不進灯預處理之情形時’直接記憶自各攝像機^ 輸入之圖像。 J下’以預處理機構19進行預處理’且圖像記憶機構12 S己憶預處理圖像之情形為例進行說明。 士姿勢特定機構14根據利用攝像機1U所得之攝像圖像(於 2中為預處理圖像),特定攝像機Ha之姿勢。此處,所謂 播係域像機之位置及方向。姿勢料機構14對攝像 機lib亦進行相同之處理。 圖3係表不利用攝像機Ua所得之攝像圖像(於本例中,根 攝像機lla進行攝像所得之圖像而生成之預處理圖像)之 :的說明圖。該攝像圖像係對固定區域93方向進行拍攝所 仵之圓像。於利用攝像機lla所得之攝像圖像中,除批料堆 〇中之相較液面25為上方之部分以外’亦拍攝有對向之側 或觀察窗2之一部分。側壁6或觀察窗2之圖像係用以特 1641J9.doc -15- 201247577 /象機之方向及位置(攝像機之姿勢)。#,形成側壁6之 , 邊界線(槽)' 上述邊界線彼此之交又部及觀察窗2 角Ρ (角隅。Ρ)係於攝像圖像内作為特徵性之圖案而出 見將如上所述之特徵性之圖案記作基準圖案。基 準圖案必需為於攝像時於同一圖像中不存在相似之圖案之 圖案。例如,若窗等之角隅之形狀、線或點等之組合成為 特徵性H财可使如上所述之組合為基準圖案。又, 如下所述,姿勢特定機構14亦可逐次更新作為基準圖案之 圓像而記憶之圖像。若攝像機之姿勢不發生變化,則基準 圓案係出現於攝像圖像内之大致固定之位置(座標)。另一方 面’若於清掃時等攝像機之姿勢發生變化,則攝像圖像内 之基準圖案之位置亦發生變化。姿勢特定機構14根據利用 攝像機113所得之攝像圖像中之基準圖案之位置,判定攝像 機iia之姿勢之偏移之有無。即,基準圖案係用以判定是否 發生攝像機之姿勢之偏移。再者’以下將表示圖像内之位 置之座標記作圖像座標。 又,就增加攝像機之姿勢偏移判定之可靠性之觀點而 言’較佳為於圖像内存在複數個基準圖案。 姿勢特定機構14記憶基準圖案之圖像及攝像圖像内之基 準圖案之圖像座標。基準圖案之圖像座標例如亦可為基準 圖案之中心位置之圖像座標。姿勢特定機構14係將例2觀 察窗2之角隅部之點21a及其周邊之圖像作為基準圖案之圖 像而記憶,並且記憶上述位置之圖像座標。將此情形時之 基準圖案之圖像之例及使用基準圖案之匹配之例示於圖4 164119.doc -16- 201247577 中。圖4(&)係表示基準圖案之圖像之例。圖叫)係表示進行 與基準圖案之匹配之攝像圖像之例。於圖4(b)中,例示與= 3相同之攝像圖像。於圖4(b)中’對與圖3所示之要素相同之 要素’標註相同之符號’並省略說明。又,於圖4⑷中為 了容易明白基準圖案’與攝像圖像相比而更大地圖示。姿 勢特定機構14係於攝像圖像與記憶之各基準圖案之圖像之 間進仃圖案匹配’特定符合記憶之各基準圖案之圖像之攝 像圖像内之部分之圖像座標。姿勢特定機構14係對上述圖 録標與記憶之圖像座標進行比較,判定攝像機I之姿勢 是否發生偏移。再者,於圖案匹配中,計算成為類似之程 度之指標值之類似度。 例如,姿勢特定機構14係於圖4⑷中例示之基準圖案之圖 像與圖4(b)所示之攝像圖像之間進行圖案匹配,特定攝像圖 像内之部分81(參照圖4(b)),特定上述部分以之圖像奸 (例如攝像圖像内之部分81之令心純)。繼而姿勢特定: 構14對上述座標與預先記憶之圖像座標進行比較,判定攝 像機1 1 a之姿勢是否發生偏移即可。 又,將攝像機之姿勢推斷中使用之特徵點記作基準點。 於基準點群中,亦可包含基準圖案内之點(例如觀察窗2之 角隅部之點21a)。於圖3中,例示將點〜2“設為基準點 之情形。姿勢特錢構14記憶基準點之圖像座標與實空間 中=基準點之3維座標作為與基準點相關之資訊。由於姿勢 特定機構14記憶「基準圖案之圖像及其圖像 盘 準點之圖像座標及3維座標」,故而可判斷圖像^基㈣ 164I19.doc •17- 201247577 案與基準點之相對位置關係。 攝像機11 a對包含基準圖案及固定區域9a之圖像進行攝 像’攝像機llb對包含基準圖案及固定區域9b之圖像進行攝 像之處理相當於圖像攝像步驟。 圖5係表示姿勢特定機構14進行之姿勢推斷動作之例之 流程圖。姿勢特定機構14係於如上所述對攝像圖像内之基 準圖案之圖像座標與記憶之圖像座標進行比較,判斷攝像 機Ha之姿勢發生偏移之情形時,使用其等圖像座標,計算 姿勢之偏移量(步驟S51)。即,姿勢特定機構14對基準圆案 於攝像圖像内以何種程度偏移進行計算。 繼而,姿勢特定機構14將攝像圖像内之基準圖案之偏移 量反映於記憶之基準點之圖像座標(步驟S52)。即,姿勢特 定機構14僅以因攝像機"3之姿勢發生偏移而攝像圖像内 之基準圖案之圖像座標發生偏移之部&,移動各基準點之 圖像座標(使基準點之圖像座標之值發生變化)。 繼而’姿勢特;t機構Μ使用上述基準點之圖像座標與實 空間中之基準點之3維座標’進行攝像機校準處理,推斷攝 像機ua之姿勢1體而言,姿勢特定機構14根據實空間中 之各基準點之3維座標算线像機&之各種姿勢中之各個 基準點之圖像座標(步驟S53)e繼而,姿勢特定機㈣將根 據各基準點之3維座標而算出之圖像座標成為最靠近如上 所述對照基㈣案之@像座標之偏移進行㈣所得之基準 點之圖像座標之座標時的姿勢判定為攝像機⑴之姿勢(步 驟 S54) 〇 164119.doc 201247577 此處,以攝像機na為例進行了說明,姿勢特定機構14亦 同樣地進行與攝像機llb相關之姿勢之偏移之有無之判定 或姿勢推斷* 圖像校準機構16根據姿勢特定機構14特定之攝像機iu 之姿勢,特定於攝像圖像内(於本例中為預處理圖像内)符合a 固定區域9a之範圍。圖6係抽出利用攝像機lu所得之攝像圖 像中符合熔解之原料之液面25之範圍的模式圖。再者,圖6 之右側及左側分別為玻璃熔融爐之上游及下游。該液面Μ 之圖像中由粗實線包圍之範圍合實空間中之固定區 域9a。圖像校準機構16根據攝像機1U之姿勢,特定並擷取 符合固定區域9a之範圍3 1 a。 其中’玻璃熔融爐内之液面之高度設為固定。該高度中 之固定區域9a之範圍係預先規定。即,固定區域\之範圍(位 置)係預先規定為實空間内之固定高度之面内之區域之位 置。因此,當攝像機lla之姿勢特定時,亦可規定利用上述 攝像機lla所得之攝像圖像内之符合固定區域9a之範圍。 即,圖像校準機構16特定將於實空間内固定高度中之固定 區域9a投影於姿勢已知之攝像機1 la之攝像圖像時之圖像内 之範圍3 1 a即可。 再者,於玻璃熔融爐内之液面之高度設為固定之情形 時,可藉由對攝像圖像中之一像素部分之偏移於實空間内 偏移幾毫米進行調查,而把握攝像圖像中之像素分辨率 (mm/pixel) 〇 又,圖像校準機構16除進行特定於圖像内符合固定區域 164119.doc 201247577 9a之範圍3 1 a之處理以外,亦進行將上述範圍3 la之圖像轉換 成自正上方(換言之,與液面對向之上方)觀察固定區域9a 時之圖像之處理。即,圖6中例示之圖像係以攝像機113之 視點(相對於液面為傾斜方向)觀察固定區域9a之情形時之 圖像’將圖像内之範圍3込轉換成使視點變化成固定區域9a 之正上方之情形時之圖像。將該轉換結果之例例示於圖7 中°如此般’圖像校準機構16對在圖像内符合固定區域9a 之範圍3 1 a進行使視點變化成固定區域9a之正上方之視點轉 換處理’生成自上述視點觀察之圖像即可。 再者’成為由圖像校準機構16轉換成自正上方觀察固定 區域9a時之圖像之對象並不限定於自利用攝像機Ua所得之 攝像圖像中擷取之範圍3 la。例如,對於藉由圖像處理(例 如’下述之背景圖像作成處理)而獲得之圖像,圖像校準機 構16亦進行相同之轉換。 圖像校準機構16對利用攝像機1 lb所得之攝像圖像(於本 例中為預處理圖像)亦進行相同之處理。 背景圖像作成機構15使用由圖像校準機構16自藉由預處 理機構19而依次生成之複數之預處理圖像中而擷取之範圍 31a(符合固定區域9a之範圍31a)之圖像,作成不存在批料堆 之情形時之液面之圖像(背景圖像作成處理)。由於該範圍 313為符合固定區域9a之圖像,故而成為以泡為背景而拍攝 有批料堆之圖像。又’由於批料堆之移動速度或炫解速度 緩慢’故而於範圍313中,始終(或以高頻率)拍攝有批料堆。 因此’作為符合固定區域9a之範圍31a,難以直接攝像僅拍 164119.doc •20· 201247577 攝有泡(背景)之狀態之圖像。因此,背景圖像作成機構15 使用自複數之圖像中擷取之範圍31a,作成不存在批料堆之 背景圖像。 於在液面中不存在批料堆之部位存在泡。又,批料堆— 面緩慢向下游方向移動一面熔解。因此,於自某圖像中榻 取之範圍313中符合批料堆之像素於自其他圖像中掏取之 範圍31a中亦表示泡。背景圖像作成機構15針對自複數之圖 像中擷取之符合固定區域9a之範圍31a中之每一對應之像素 之組(換言之,針對符合固定區域9(1内之相同位置之每一像 素之組),特定符合泡之亮度,藉此作成不存在批料堆而僅 表示有批料堆之背景之圖像》再者,此處,以針對自複數 之圖像中擷取之範圍31a中之每一對應像素之組進行處理 之情形為例,但亦可針對自複數之圖像中擷取之範圍3込中 之每-對應之區域’特定符合泡之亮度。區域係連續之像 素聚集而形成之區域β 背景圖像作成機構15對利用攝像機Ub所得之攝像圖像 (於本例中為預處理圖像)亦進行相同之處理。 、差分運算機構17計算2張圖像間之對應之像素間之差 分。具體而言’自拍攝有批料堆之圖像之各像素之亮度值 減去背景圖像中之對應像素之亮度值。藉由該減法處理, 獲得自拍攝有批料堆之圖像中將背景部分去除之圖像。惟 泡之亮度亦多少會有變化。因此,自拍攝有批料堆之圖像 哀符口 /包之像素之亮度值減去背景圖像中之對應像素之 儿度值所得之結果未必為〇。因此,差分運算機構17較佳為 164] J9.doc •21 · 201247577 =自拍攝有批料堆之圖像之各像素之亮度值減去背景圖像 中之對應像素之亮度值後,進行將每—像素之減法結果二 值化為0」4「1」之處理。於該二值化處理中,差分運 算機構17針對每—像素,若減法結果為特定值以上,則將 減法結果進位至M」,若減法結果未達上述特;t值,則將 減法結果降位至「〇」即可。藉由進行該二值化處理,可更 明確地區別符合批料堆之區域(亮度值為「1」之區域)與符 合背景之區域(亮度值為「0」之區域)。 觀察資料算出機構18根據將背景部分去除而符合批料堆 之部分所剩餘之圓像’算出批料堆之觀察資料。作為觀察 資料之例,可列舉例如批料堆之頂端位置、批料堆之移動 速度、批料堆之熔解速度(批料堆之減少率)、固定區域 9b各自中之批料堆之佔有率等。又,關於該等觀察資料, 亦可算出固定區域9a中之值與固定區域9b中之值之差,而將 上述差設為觀察資料。 又,亦可將固定區域9a平分成側壁側之區域與玻璃熔融 爐之寬度方向之中央側之區域,將上述兩個區域中之批料 堆之佔有率之比(以下記作内外比)作為觀察資料而進行計 算。同樣地,關於固定區域9b,亦可平分成側壁側之區域 與玻璃熔融爐之内側之區域,將上述兩個區域中之批料堆 之佔有率之比(内外比)作為觀察資料而進行計算。 預處理機構19'姿勢特定機構14、背景圖像作成機構15、 圖像校準機構16、差分運算機構π及觀察資料算出機構18 係藉由例如依據程式而進行動作之電腦之CPU(centrai 164119.doc •22· 201247577It can also be the following method. Gan & A, , and package 3 convert the background image to the background image of the circle image when the surface of the image is fixed from the upper surface of the liquid to the upper surface of the surface. In the observation data calculation, the observation data is calculated based on the background exclusion image after the background image conversion using the background exclusion. Like the photoshooting method, the method includes the following steps: the preprocessing step is performed, which is to obtain the roundness of the brightness and the darkness in the image obtained by the image obtained in the image capturing step: The predetermined amount of contrast may be a method in which, in the pre-processing step, the number of edges in the image is calculated as the amount indicating the contrast, and a plurality of graphs satisfying the condition that the number of edges is equal to or greater than a predetermined % value is selected. An image of an object that becomes (4) an area conforming to the range of (4) is generated as 'based on the selected plural image'. Further, the glass melting furnace operating method of the present invention is characterized in that it comprises the following step of "influence degree deriving step of extracting the operating parameters of the glass melting furnace for calculating the observation data in the monitoring method of the glass melting furnace; The degree of influence given by the observation data; and the melting furnace control step is an operation parameter in which the absolute value of the degree of influence on the observation data is changed to a predetermined value or more when the observation data satisfies a specific condition. Further, the in-glass monitoring system of the present invention is characterized in that it includes a mechanism for an image pickup mechanism that includes a reference pattern provided in a glass melting furnace and a liquid surface of a glass material melted in a glass melting furnace. The image of the fixed range is imaged; the image calibration mechanism is calculated according to the use of the image; the image capturing mechanism calculated by the positional deviation of the reference pattern in the image 1641l9.doc 201247577 =!: the image taken from the image The inside captures a region that conforms to a fixed range; the back: the image forming mechanism's a plurality of plum-shaped round images that are extracted from the image that is self-recovering as a region that conforms to a fixed range, and is formed into a glass: in a melting furnace The glass material is the background image of the background of the batch pile; the differential mechanism 'by the pixels of the image of the couch taken out from the image being imaged as a region conforming to the fixed range for each pixel The brightness value is subtracted from the brightness value of the corresponding pixel in the background image, and the background image except the background except the background in the image captured with the batch pile and the #view is generated. And the observation data calculation mechanism calculates the observation data related to the batch pile based on the background exclusion image. It may also be configured as follows: the background image creating mechanism counts the number of pixels that match each brightness value for each corresponding pixel or each corresponding area of the captured image of the plurality, and counts the pixels according to the respective brightness values. As a result, the brightness value representing the background is determined, thereby creating a background circle image. The difference calculation mechanism may be configured to: for each pixel, perform a luminance value of a pixel of the captured image as a region conforming to the fixed range from the captured image, and subtract the corresponding pixel in the background image. The processing of the luminance value 'binarizes the subtraction result of each pixel to generate a background exclusion image' may also be configured as follows: the image calibration mechanism converts the background image into a liquid-facing surface When the image in the fixed range is viewed from above, the captured image captured as a region conforming to the fixed range is converted into an image when the fixed range is viewed from above the liquid facing surface, and the difference computing mechanism performs self-utilization. The brightness value of the captured image after the conversion by the calibration mechanism is subtracted from the brightness value of the corresponding pixel in the background image converted by the calibration mechanism of the image 164H9.doc 201247577, and the observation data calculation mechanism is based on the difference The background exclusion image generated by the calculation mechanism calculates the observation data. The image calibration mechanism may convert the background exclusion image generated by the difference calculation mechanism into an image when the fixed range is viewed from above the liquid facing direction, and the observation data calculation mechanism according to the utilization image. The calibration mechanism calculates the observation data by the background image after the conversion. The configuration may include a preprocessing mechanism that calculates the amount of contrast between the brightness and the darkness in the image for each image obtained by the image capturing means, and selects that the amount of contrast is predetermined. An image of the condition. The configuration may be such that the preprocessing unit calculates the number of edges in the image as the amount indicating the contrast, and selects an image that satisfies the condition that the number of edges is equal to or greater than a predetermined threshold, and based on the selected plural image, Generates an image that is an object that captures a region that fits a fixed range. It may be a configuration including an observation data analysis unit that indicates the degree of influence of the operation parameters of the glass melting furnace on the observation data calculated by the observation data calculation means. Further, it may be a configuration including a melting furnace control unit that changes the absolute value of the degree of influence on the observation data to a predetermined value or more when the data is observed to satisfy a specific condition. Further, the method for producing a glass article of the present invention is characterized in that it comprises the steps of: a glass melting step of producing a molten glass in a glass melting furnace; and a clarifying step of removing a bubble of the molten glass in the clarification tank; 'The molten glass that is formed by foaming; and the slow cooling step' I64119.doc • 10·201247577 The slow-cooling of the formed molten glass; and the following steps are included: the mechanism is included in the glass _ The image of the fixed range of the liquid surface of the glass material in the rail melting furnace is imaged; the step of taking the reference image in the image is based on the posture of the structure, and the calculation is performed according to the posture of the image. The image of the image camera captures a region that conforms to a fixed range. The background image creation step is based on a plurality of extracted circular images captured as a region that conforms to a fixed range, and is created in a scene (4). The back of the step _; the back subtracts the processing of the brightness value of the corresponding pixel in the background image, and generates the state from the batch with the batch and the background: the background image except the background And the observation data calculation step, which calculates the observation data related to the batch pile according to the background exclusion image. [Effects of the Invention]: According to the monitoring method and the visual system of the glass smelting furnace of the present invention, the observation of the fixed area in the glass melting furnace is continued, and the state of the batch of the batch material is... according to the manufacture of the glass mouth The method 'can be achieved like! _ Road Glass-making H material is in good condition and the operating method of the glass-cooling furnace in the second and second inventions can make the furnace-running parameters clear. 164] 19.doc -11 - 201247577 Hereinafter, embodiments of the present invention will be described with reference to the drawings. First, an example of a glass melting furnace to which the glass melting furnace monitoring system of the present invention is applied will be described. The figure shows a plan view of an example of the glass melting furnace as described above. The glass melting furnace is made of a glass material which is surrounded by the bottom surface, the upstream wall (the upstream side wall) 7, the side wall 6, the downstream wall (the downstream side wall) 8, and the ceiling (not shown). Dissolved. The upstream wall 7 is provided with input ports 3a and 3b for inputting raw materials, and the downstream wall 8 is provided with a discharge port 4 for discharging the molten glass raw material. Further, on the side wall 6, an observation window 2 and a burner 5 are provided, respectively. Fig. 1 shows the case where the inlet ports 3a and 3 (} are provided, but the number of the inlet ports is not limited to two. The glass raw materials in the solid state are supplied from the inlet ports 3a and 3b. The fire blown by the burner 5 is heated, so that the raw material is slowly melted, and the molten raw material is slowly moved to the downstream side and discharged from the discharge port 4. The raw materials stacked in the solid state in the glass melting furnace 1 are piled up in batches. The material stacking system is melted while moving toward the downstream side as time passes. The glass melting furnace monitoring system of the present invention includes a camera 11 {1, 11 (), and a fixed area 9a for monitoring the liquid level in the glass melting furnace, 9b. In Fig. 1, the case where two fixed areas 9a and \ are defined so as to cover the area between the side walls in the front direction of each camera in the liquid level in the furnace is covered by the two fixed areas 9a and 9b. The 118-series is imaged from the upstream side, and the fixed area 9a on the right side (hereinafter referred to as only the fixed area 9a) is imaged, and the camera 11b is imaged from the upstream side to the fixed area 9b on the left side (hereinafter simply referred to as the fixed area 9b). In the present invention, the case where the glass melting furnace monitoring system includes two cameras 11a and 11b will be described as an example, but the number of cameras included in the glass melting furnace monitoring system 164119.doc 12 201247577 is not limited to 2. Further, the solid-solid regions 9a and 9 are separated from the vicinity of the input ports 3a and 3b, and are defined as follows: when the image is taken as a fixed region in the vicinity of the casters Wa and 3b. The part of the captured image that conforms to the fixed area becomes completely the batch pile. The bubble that becomes the background is not likely to be photographed. In the above case, the data related to the batch pile cannot be calculated. [Embodiment 1] 2 is a block diagram showing a configuration example of the glass-melting furnace monitoring system according to the first embodiment of the present invention. The glass-melting furnace monitoring system according to the first embodiment includes a camera 11a, a camera 11b, and an image processing device 13. The in-furnace monitoring system performs the same processing on the images obtained by imaging the cameras 11a and (1), respectively. Therefore, the description of the camera 11a will be omitted as appropriate with the camera 11b. The camera 11a repeats the image of the fixed area 9a of the liquid level through the observation window 2 of the glass melting furnace (see FIG. 。). The image is a still image. Similarly, the camera 11b is also passed through the glass melting furnace. Observation window 2 (refer to the figure υ, repeating the still image of the fixed area 9b of the imaging liquid surface. The imaging interval of the cameras i丨&, 丨丨b may be predetermined. Further, the imaging range of the camera 11a (the field of view) The range is not only the income fixed area 9a, but also the liquid level in the vicinity of the fixed area\ or the side wall opposite to the camera Ua. Therefore, in the captured image of the camera 11a, the solid area 9a and the liquid in the vicinity thereof are also photographed. Face or opposite side wall. The same applies to camera 1. The images captured by the cameras 11a, 11b are input to the image processing device i3. 164119.doc 5: •13· 201247577 Picture:=Set 13 pairs of images taken by the camera Ua image = a variety of data related to the batch pile in the fixed area 9a (eg related to - set or action) Information). Similarly, the image processing apparatus 13 performs image processing on the image captured by the camera 1U, and calculates a plurality of types of data related to the pile in the fixed area. Hereinafter, the data of the batch pile calculated based on the images taken by the cameras 11a and 11b is recorded as observation data. The image processing device 13 includes a preprocessing unit 19, an image memory unit 12, a posture specifying unit 14, a background image forming unit 15, an image matching unit 16, a difference calculating unit 17, and an observation data calculating unit 18. The pre-processing unit 19 generates an image in which the raw material powder or the flame (the fire ejected from the burner 5) is not imaged based on the image captured by the camera 11a. If! When the original powder or flame floating in the glass melting furnace is shot in the round image, the image of the batch pile becomes unclear. The preprocessing unit 19 uses the plurality of images obtained by imaging by the camera 1U to generate an image in which the state of the batch pile is clearly captured without being affected by disturbances such as raw material powder or flame. The pre-processing unit 19 performs the same processing on the image obtained by imaging the camera iib. An image obtained by removing the influence of the raw material powder or the flame is recorded as a pretreatment. Further, there is a case where an image generated by the preprocessing unit 19 based on a plurality of images captured by a camera is recorded as a preprocessed image. The 'pre-processed image is the same as the image obtained by each camera except that the influence of the raw material powder or the flame is removed to make the batch heap clearer, and the pre-processed image is only recorded as a camera. The situation of the image. In other words, there is a case where the image itself obtained by imaging with the camera is referred to as a captured image. The pre-processing mechanism 19 respectively obtains the pre-processed 164119.doc -14·201247577 image obtained according to the camera 1丨3 and the pre-processed image memory memorizing mechanism 12 obtained according to the camera 11b. A situation where pre-processing or a part is not required. For example, in a glass melting furnace where the influence of the flame is small or the suspended raw material powder is less, the pretreatment may not be performed. In the above situation, the county 13 of the image can directly recall the image input from the image camera Ub to the image memory mechanism 12. The image memory mechanism (10) memorizes the memory of the image. As described above, the pre-processing unit 19 memorizes the pre-processed image obtained by the above-described pre-processing when pre-processing the image input from each camera lu. Also, ^ directly enters the image input from each camera ^ when the lamp is not preprocessed. The case where J is 'preprocessed by the preprocessing mechanism 19' and the image memory mechanism 12 has recalled the preprocessed image will be described as an example. The pose specific mechanism 14 specifies the posture of the camera Ha based on the captured image obtained by the camera 1U (preprocessed image in 2). Here, the location and direction of the broadcast system camera. The posture material mechanism 14 also performs the same processing on the camera lib. Fig. 3 is an explanatory diagram showing a captured image obtained by the camera Ua (in this example, a preprocessed image generated by imaging the image obtained by the camera 11a). This captured image is a circular image captured in the direction of the fixed area 93. In the captured image obtained by the camera 11a, in addition to the portion of the batch stack which is higher than the liquid surface 25, a portion opposite to the side or the observation window 2 is also photographed. The image of the side wall 6 or the observation window 2 is used for the direction and position of the camera (camera posture) of 1641J9.doc -15-201247577/camera. #, forming the side wall 6, the boundary line (slot)' The boundary line between the above-mentioned boundary lines and the observation window 2 angle 隅 (angle 隅.Ρ) is shown in the captured image as a characteristic pattern. The characteristic pattern described is referred to as a reference pattern. The reference pattern must be such that there is no pattern of similar patterns in the same image at the time of imaging. For example, a combination of a shape, a line, a dot, or the like of a corner or the like of a window or the like can be combined as a reference pattern as described above. Further, as will be described later, the posture specifying unit 14 may sequentially update the image memorized as the circular image of the reference pattern. If the posture of the camera does not change, the reference circle appears at a substantially fixed position (coordinate) in the captured image. On the other hand, if the posture of the camera changes during cleaning, the position of the reference pattern in the captured image also changes. The posture specifying unit 14 determines the presence or absence of the deviation of the posture of the camera iia based on the position of the reference pattern in the captured image obtained by the camera 113. That is, the reference pattern is used to determine whether or not the shift of the pose of the camera occurs. Further, the following indicates that the position of the position in the image is marked as an image coordinate. Further, from the viewpoint of increasing the reliability of the posture shift determination of the camera, it is preferable that a plurality of reference patterns exist in the image. The posture specifying unit 14 memorizes the image of the reference pattern and the image coordinates of the reference pattern in the captured image. The image coordinates of the reference pattern may, for example, also be image coordinates of the center position of the reference pattern. The posture specifying mechanism 14 stores the image of the point 21a of the corner portion of the observation window 2 and its surroundings as an image of the reference pattern, and memorizes the image coordinates of the above position. An example of the matching of the image of the reference pattern and the use of the reference pattern in this case is shown in Fig. 4 164119.doc -16-201247577. Fig. 4 (&) shows an example of an image of a reference pattern. The figure is an example of a captured image that matches the reference pattern. In FIG. 4(b), the same captured image as =3 is exemplified. In Fig. 4(b), the same elements as those shown in Fig. 3 are denoted by the same reference numerals, and the description thereof will be omitted. Further, in Fig. 4 (4), it is easy to understand that the reference pattern ' is larger than the captured image. The posture-specific mechanism 14 is an image coordinate of a portion of the image of the image corresponding to each of the reference patterns of the memory that is matched between the captured image and the image of each of the reference patterns of the memory. The posture specifying unit 14 compares the above-mentioned image label with the image coordinates of the memory, and determines whether or not the posture of the camera 1 is shifted. Furthermore, in the pattern matching, the similarity of the index values which are similar degrees is calculated. For example, the posture specifying mechanism 14 performs pattern matching between the image of the reference pattern illustrated in FIG. 4 (4) and the captured image shown in FIG. 4 (b), and the portion 81 in the specific captured image (refer to FIG. 4 (b) )), the specific part of the above image is used for imagery (for example, the part 81 of the captured image is pure). Then, the posture is specified: The configuration 14 compares the coordinates with the image coordinates that are memorized in advance, and determines whether or not the posture of the camera 1 1 a is shifted. Further, the feature points used in the estimation of the posture of the camera are referred to as reference points. The reference point group may also include a point in the reference pattern (e.g., point 21a of the corner of the observation window 2). In Fig. 3, the case where the dot ~ 2 " is set as the reference point is exemplified. The image coordinates of the memory reference point and the 3 -dimensional coordinate of the reference point in the real space are used as information related to the reference point. The posture-specific mechanism 14 memorizes the "image coordinates of the reference pattern and the image coordinates of the image disc and the three-dimensional coordinates", so that it is possible to determine the relative positional relationship between the image and the reference point of the image (4) 164I19.doc • 17-201247577 . The camera 11a images the image including the reference pattern and the fixed area 9a. The processing by which the camera 11b images the image including the reference pattern and the fixed area 9b corresponds to an image capturing step. Fig. 5 is a flow chart showing an example of the gesture estimation operation performed by the posture specifying unit 14. The posture specifying unit 14 compares the image coordinates of the reference pattern in the captured image with the image coordinates of the memory as described above, and determines that the posture of the camera Ha is shifted, and uses the image coordinates to calculate The amount of shift of the posture (step S51). That is, the posture specifying unit 14 calculates the degree of deviation of the reference circle in the captured image. Then, the posture specifying unit 14 reflects the offset amount of the reference pattern in the captured image on the image coordinates of the reference point of the memory (step S52). In other words, the posture specifying unit 14 shifts the image coordinates of the reference points of the reference image in the captured image by the shift of the posture of the camera "3, and moves the image coordinates of the reference points (the reference point is made) The value of the image coordinates changes). Then, the 'pose position; t mechanism Μ uses the image coordinate of the reference point and the 3D coordinate of the reference point in the real space to perform the camera calibration process, and the posture of the camera ua is estimated. The posture specific mechanism 14 is based on the real space. The image coordinates of each of the various positions in the three-dimensional coordinate line camera of each of the reference points (step S53) e are followed by the posture specific machine (four) calculated based on the three-dimensional coordinates of each reference point. The image coordinate is determined to be the closest to the coordinate of the image coordinate of the reference point obtained by the (4) offset of the control base (4) as described above, and the posture of the camera (1) is determined as the posture of the camera (1) (step S54) 〇 164119.doc 201247577 Here, the camera na has been described as an example, and the posture specifying unit 14 similarly performs the determination of the deviation of the posture associated with the camera 11b or the posture estimation. * The camera of the image specifying mechanism 16 based on the posture specifying mechanism 14 The posture of iu is specific to the range of the fixed area 9a in the captured image (in this example, the preprocessed image). Fig. 6 is a schematic view showing the range of the liquid level 25 of the material in accordance with the melting in the image obtained by the camera lu. Furthermore, the right side and the left side of Fig. 6 are the upstream and downstream of the glass melting furnace, respectively. The area surrounded by the thick solid line in the image of the liquid level 合 is in the fixed area 9a in the solid space. The image calibration mechanism 16 specifies and captures a range 3 1 a that conforms to the fixed area 9a in accordance with the posture of the camera 1U. The height of the liquid level in the glass melting furnace is fixed. The range of the fixed area 9a in the height is predetermined. That is, the range (position) of the fixed area \ is defined as the position of the area within the plane of the fixed height in the real space. Therefore, when the posture of the camera 11a is specified, the range of the fixed area 9a in the captured image obtained by the above-described camera 11a can be specified. That is, the image aligning means 16 specifies that the fixed area 9a of the fixed height in the real space is projected in the range of the image 3 1 a in the image at the time of capturing the image of the camera 1 la having the known posture. Furthermore, when the height of the liquid surface in the glass melting furnace is set to be fixed, the image can be captured by investigating the offset of one pixel portion of the captured image in the real space by a few millimeters. In the pixel resolution (mm/pixel) in the image, the image calibration mechanism 16 performs the above range 3 la in addition to the processing in the image that is specific to the range of the fixed region 164119.doc 201247577 9a 3 1 a. The image is converted into an image of the image when the fixed area 9a is viewed from directly above (in other words, above the liquid facing direction). That is, the image illustrated in Fig. 6 is an image in the case where the fixed region 9a is observed from the viewpoint of the camera 113 (the oblique direction with respect to the liquid surface). The range within the image is converted to a point where the viewpoint is changed to be fixed. An image of the situation immediately above the area 9a. An example of the result of the conversion is shown in Fig. 7. The image-aligning mechanism 16 performs a viewpoint conversion process of changing the viewpoint to directly above the fixed region 9a in the range 3 1 a in the image conforming to the fixed region 9a. The image observed from the above viewpoint can be generated. Further, the object which is converted into the image when the image matching mechanism 16 is converted into the fixed region 9a from the upper side is not limited to the range 3 la extracted from the captured image obtained by the camera Ua. For example, the image calibration mechanism 16 performs the same conversion on the image obtained by image processing (e.g., the background image creation processing described below). The image calibration mechanism 16 performs the same processing on the captured image (preprocessed image in this example) obtained by the camera 1 lb. The background image creating unit 15 uses an image of a range 31a (corresponding to the range 31a of the fixed area 9a) captured by the image calibration unit 16 from a plurality of preprocessed images sequentially generated by the preprocessing unit 19, An image of the liquid surface when the batch is not present (background image creation processing). Since the range 313 is an image conforming to the fixed area 9a, an image of the batch pile is taken with the bubble as the background. In addition, in the range 313, the batch pile is always taken (or at a high frequency) because of the moving speed of the batch pile or the slow speed of the scattering. Therefore, as the range 31a conforming to the fixed area 9a, it is difficult to directly capture an image in which only 164119.doc • 20·201247577 has a bubble (background). Therefore, the background image creating means 15 uses the range 31a extracted from the complex image to create a background image in which the batch pile is not present. There is a bubble in the portion of the liquid surface where the batch is not present. In addition, the batch pile is slowly melted in the downstream direction. Therefore, the pixels in the range 313 of the image taken from a certain image in the range 31a taken from the other images also represent bubbles. The background image creating unit 15 is for the group of pixels corresponding to each of the ranges 31a of the fixed area 9a captured from the complex image (in other words, for each pixel in the same position in the fixed area 9) Group), specifically meets the brightness of the bubble, thereby creating an image without the batch pile and only representing the background of the batch pile. Again, here, the range 31a taken from the image of the complex number For example, the case where each group of corresponding pixels is processed, but it is also possible to specifically match the brightness of the bubble for each of the range 3's selected from the image of the complex number. The region is continuous pixels. The area formed by the aggregation β background image creating means 15 performs the same processing on the captured image obtained by the camera Ub (preprocessed image in this example). The difference computing means 17 calculates between the two images. Corresponding pixel-to-pixel difference. Specifically, the brightness value of each pixel in the image from the image of the batch is subtracted from the brightness value of the corresponding pixel in the background image. With this subtraction process, the self-shooting batch is obtained. Image of pile The image is removed from the background. The brightness of the bubble will also change somewhat. Therefore, the brightness value of the image of the image of the batch of the sorrow/package is subtracted from the corresponding pixel in the background image. The result obtained by the degree value is not necessarily 〇. Therefore, the difference operation mechanism 17 is preferably 164] J9.doc • 21 · 201247577 = the brightness value of each pixel from the image of the batch pile is subtracted from the background image After the luminance value of the pixel is corresponding, a process of binarizing the subtraction result of each pixel into 0"4"1" is performed. In the binarization processing, the difference calculation means 17 specifies the subtraction result for each pixel. If the value is above, the result of the subtraction is rounded up to M". If the result of the subtraction does not reach the above special value; the value of t is reduced to "〇". By performing the binarization process, the difference can be more clearly distinguished. The area corresponding to the batch pile (the area where the brightness value is "1") and the area corresponding to the background (the area where the brightness value is "0"). The observation data calculation means 18 conforms to the portion of the batch pile according to the removal of the background portion. The remaining circle is like 'calculated batch Observation data. As an example of observation data, for example, the top position of the batch pile, the moving speed of the batch pile, the melting speed of the batch pile (reduction rate of the batch pile), and the batch in each of the fixed areas 9b can be cited. In addition, as for the observation data, the difference between the value in the fixed region 9a and the value in the fixed region 9b can be calculated, and the difference can be used as the observation data. Further, the fixed region 9a can also be used. The area which is divided into the side of the side wall side and the area on the center side in the width direction of the glass melting furnace is calculated by taking the ratio of the occupation ratio of the batch piles in the above two areas (hereinafter referred to as the internal/external ratio) as observation data. The fixed region 9b may be divided into a region on the side wall side and a region on the inner side of the glass melting furnace, and the ratio of the occupancy ratio of the batch piles in the above two regions (inside-outside ratio) is calculated as observation data. The pre-processing mechanism 19' posture specifying means 14, the background image forming means 15, the image correcting means 16, the difference calculating means π, and the observation data calculating means 18 are CPUs of a computer operated by, for example, a program (centrai 164119. Doc •22· 201247577
Processing Unit ’中央處理單元)而實現^於此情形時, 讀入例如記憶於電腦之程式記憶裝置(省略圖示)中之程 式,CPU依據上述程式,作為預處理機構19、姿勢特定機 構I4、背景圖像作成機構15、圖像校準機構16、差分運算 機構17及觀察資料算出機構18進行動作即可。 繼而’對動作進行說明。 首先,對利用預處理機構19之預處理進行說明。攝像機 na定期地對固定區域9a方向進行攝像,絲錢像依次輸 入至預處理機構19。預處理機構19係針對每—固定之週期 (例如數秒之週期根據於上述内自攝像機1W輸入之 複數之圖像,生成預處理圖像。具體而言,預處理機構19 係對在m期内輸入之各個圖像,計數圖像内之邊緣數。再 者’所謂邊緣’係指出現於圖像内之線。亦可將設為圖像 内之邊緣數之計數對象之區域,限定於例如相當於壁面之 區域及相當於較區域9a之區域1用預處理機構19之處 理週期較短’於在上述週期内自攝像機⑴輸人之各圖像 中,所拍攝之批料堆之數量之多少不發生變化之情形較 多。又,拍攝於圖像中之批料堆之多少不發生變化是指若 不存在火焰或補粉之邊絲亦絲持某程度之 多少。制上述情況,預處理機構19係自於i週期内自攝像 機ua輸入之複數之圖像中’選擇保持邊緣數之計數結果較 多之狀態之連續之複數之圖推 ^ ^ 之圖像。再者,作為判斷圖像内之 邊緣數之計數結果之多少之基準,亦可使用例如預先規定 之閾值#體而έ,預處理機構19係於滿足作為計數結果 164119.doc -23. 201247577 而獲得之邊緣數為對邊緣數預先規定之間值以上之條件之 情形時,判定圖像内之邊緣數較多,而選擇邊緣數為閾值 以上之圖像即可。又,預處理機構19係於作為計數結果而 獲得之邊緣數未達閾值之情形時,判定圖像内之邊緣數較 少,而不選擇邊緣數未達閾值之圖像,或,亦可根據所輸 入之各圖像中之邊緣數之計數結果,變動邊緣數之多少之 判斷基準。 又,於上述說明中,以預處理機構19選擇連續之複數之 圖像之情形為例進行了說明,但預處理機構19選擇之複數 之圖像亦可不為連續之圖像。 又,預處理機構19算出表示圖像内之明暗之對比度之 之條件之圖 選擇滿足對表示上述對比度之量預先規定 比度之量預 用基於邊緣 於以下。例, 每一圖像, 包含之各像. 規定拍攝有4 出圓像内之_ 擇圖像之條{ 之對比度之童相較前一 像即可上述之邊緣數係表示圖像内之明暗之對比度之量 之一例。又,邊綾齪蛊朗佶w μ 4』,丄 對比度之量In the case where the processing unit 'central processing unit' is implemented, a program stored in a program memory device (not shown) of the computer is read, for example, the CPU is used as the preprocessing unit 19 and the posture specifying unit I4 according to the program. The background image creating unit 15, the image matching unit 16, the difference calculating unit 17, and the observation data calculating unit 18 may operate. Then, the action will be described. First, the preprocessing using the preprocessing mechanism 19 will be described. The camera na periodically images the fixed area 9a, and the silk image is sequentially input to the preprocessing mechanism 19. The pre-processing mechanism 19 generates a pre-processed image for each fixed period (for example, a period of several seconds according to the above-described complex image input from the camera 1W. Specifically, the pre-processing mechanism 19 is in the m period Each image is input, and the number of edges in the image is counted. Further, 'the so-called edge' refers to a line appearing in the image. The area of the counting object set as the number of edges in the image may be limited to, for example, The area corresponding to the wall surface and the area 1 corresponding to the area 9a are shorter by the processing period of the preprocessing mechanism 19 'the number of the batch piles taken in each of the images input from the camera (1) in the above period There are many cases where there is no change. Moreover, the number of batches taken in the image does not change. It means that if there is no flame or powder, the wire will be held to a certain extent. The processing unit 19 is an image in which a plurality of consecutive numbers of the states in which the number of counts of the number of edges is counted are selected from the plurality of images input from the camera ua in the i-cycle. Like the inner edge The basis of the number of counting results may be, for example, a predetermined threshold value, and the preprocessing mechanism 19 is configured to satisfy the number of edges obtained as the counting result 164119.doc -23. 201247577 as the number of edges is predetermined In the case of a condition above the value, it is determined that the number of edges in the image is large, and the image whose edge number is equal to or greater than the threshold value is selected. Further, the preprocessing mechanism 19 is based on the number of edges obtained as a result of the counting. When the threshold is reached, it is determined that the number of edges in the image is small, and the image whose edge number does not reach the threshold is not selected, or the number of edges may be changed according to the counting result of the number of edges in each image input. Further, in the above description, the case where the preprocessing unit 19 selects a continuous plurality of images has been described as an example. However, the image selected by the preprocessing unit 19 may not be a continuous image. Further, the preprocessing unit 19 calculates a map indicating the condition of the contrast between the brightness and the darkness in the image, and satisfies the amount of the pre-specified ratio for the amount of the contrast. Next, for example, each image, including each image. The specified shooting has 4 out of the circle image. The contrast image of the child is the same as the previous image. An example of the amount of contrast between light and dark. In addition, the side is 佶w μ 4』, the amount of contrast
相較前—個表像之對比度之量降低 164119.doc •24· 201247577 值以上之現象發生時直至上述固定時間經過後為止之圖像 除外,選擇未除外而剩餘之圖像。例如,於採用該條件, 算出亮度值之標準偏差作為表示明暗之對比度之量之情形 時,預處理機構19係於在某圖像中,亮度值之標準偏差相 較之前之圖像之亮度值之標準偏差降低固定值以上之情形 時,將自上述時間點起直至經過固定期間為止所生成之圖 像自其後之處理對象中除外,選擇未除外而剩餘之圖像即 可繼而,預處理機構19根據所選擇之複數之圖像生成預 處理圖像。再者,表示圖像内之明暗之對比度之量降低固 疋值以上係指對比度急遽降低,可視為產生原料粉飛揚等 現象。 於以下之說明中’以預處理機構19根據圖像内之邊緣數 選擇圖像之情形為例進行說明。 預處理機構19使用所選擇之複數之圖像,決定預處理圖 像中之各個像素之亮度值,藉此生成預處理圖像。於所選 擇之複數之圖像中,著眼於對應之像素(相同之圖像座標之 像素)特疋上述像素令成為最小之亮度值。繼而,預處理 機構19將上述冗度值規定為預處理圖像中之對應像素之亮 又值例如帛處理機構19讀入所選擇之各圖像中之圖像 座標(Xl ’ yi)之亮度值,特定圖像座標(X丨,y,)中之亮度值 最J值繼而,預處理機構19將上述成為最小之亮度 值規定為預處理圖像之圖像座標(χι,_之亮度值。預處 機構19針對每—像素進行該處理°繼而,預處理機構19 將所生成之預處理圖像記憶於圖像記憶機構12中。預處理 164119.doc -25· 201247577 機構19係以固定週期反覆該處理。因此,將根據攝像機1込 進行攝像所得之圖像而生成之預處理圖像依次儲存於圆像 記憶機構12中。 再者’於預處理中,對於自攝像機lla輸入之複數之圖像 中「保持邊緣之計數結果較多之狀態之連續之複數之圖像」 以外之圖像,忽視即可。 此處’以使用攝像機lla進行攝像所得之圖像之情形為例 進行了說明’攝像機llb亦定期地對固定區域9b方向進行攝 像’並將其圖像依次輸入至預處理機構19。預處理機構19 亦根據攝像機11 b進行攝像所得之圖像,同樣地生成預處理 圖像’並使其記憶於圖像記憶機構12中。 可以說保持邊緣之計數結果較多之狀態之連續之複數之 圖像為不怎麼拍攝有火焰或原料粉之圖像。其原因在於: 於拍攝有較多之火焰或懸浮之原料粉之圖像中,批料堆或 側壁變得不清楚,而圖像内之邊緣數減少。又,於拍攝有 火焰之情形時,於圖像内符合火焰之部位之亮度值變為較 同之值。因此,如上所述,選擇複數個不怎麼拍攝有火焰 或原料粉之圖像,進而特定其等圖像中之對應像素中最小 之亮度值,藉此可選擇未拍攝有火焰或原料粉之狀態之圖 像中之亮度值。由於規定預處理圖像作為具有如上所述之 亮度值之圖像,故而即便於攝像機lla進行攝像所得之圖像 之一部分中拍攝有於爐内懸浮之原料粉或火焰,亦可生成 將如上所述之原料粉或火焰排除之預處理圖像。即可獲 得清楚地拍攝有設為監視對象之批料堆之圖像。預處理機 164119.doc • 26 · 201247577 構19生成預處理圖像之動作相當於預處理步驟β 再者,如以上說明般,於火焰之影響較小,或懸浮之原 料粉較少之玻璃熔融爐中’無需進行如上所述之預處理。 於上述之情形時,圖像處理裝置13將攝像機丨匕、lu進行攝 像所得之圖像直接記憶於圖像記憶機構丨2中即可。 繼而,對姿勢特定機構14判斷攝像機之姿勢之動作進行 說明。此處,以判斷攝像機lla之姿勢之情形為例,攝像機 1U之姿勢判斷處理亦相同。圖8係表示攝像機之姿勢判斷 處理之處理經過之例之流程圖。再者,於本例中,以姿勢 特定機構14記憶複數之基準圖案之圖像及其圖像座標之情 形為例進行說明。 如上所述,預處理機構19係針對每一固定之週期(例如數 秒之週期),根據攝像機11 a進行攝像所得之圖像生成預處理 圖像,並將上述圖像記憶於圖像記憶機構12中。 姿勢特定機構14讀入記憶於圖像記憶機構12令之複數之 攝像圖像(於本例中,為根據攝像機u &進行攝像所得之圖像 而生成之預處理圖像定期地進行判斷攝像機丨、之姿勢是 否發生偏移之處理。其中,與預處理機構19之處理週期例 如為數秒相比,姿勢特定機構14之處理週期長於利用預處 理機構19之處理週期。例如,姿勢特定機構14之處理週期 亦可設為數小時。 姿勢特定機構14係當判斷成為處理開始時序時,讀入記 隐於圖像δ己憶機構12中之較近之特定張數之攝像圖像(根 據攝像機lla進行攝像所得之圖像而生成之預處理圖像該 164119.doc -27· 201247577 特定張數預先規定即可。姿勢特定機構14生成所讀入之較 近之特定張數之攝像圖像(預處理圖像)之平均圖像(步驟 S1)»具體而言,姿勢特定機構14係對所讀入之特定張數之 攝像圖像,針對每一對應之像素計算亮度值之平均值,生 成將上述平均值設為亮度值之圖像,將上述圖像設為平均 圖像即可。於本例中,例示生成平均圖像之情形,但亦可 針對每一對應之像素計算亮度值之中間值,生成將上述中 間值設為亮度值之圖像(中間值圖像)。 又’於本例中’例示於步驟S1中根據複數之圖像生成平 均圖像之情形’但亦可對記憶於圖像記憶機構12中之1張圖 像進行步驟S2以後之處理。即,亦可省略步驟§丨之處理。 姿勢特定機構14對在步驟S1中生成之平均圖像,進行與 姿勢特定機構14預先記憶之複數之基準圖案相關之圖案匹 配(步驟S2) »於本例中,以圖像彼此類似之程度越高,所 計算之類似度之值越小之情形為例進行說明。於步驟S2 中’姿勢特定機構14係於預先記憶之基準圖案之圖像與平 均圖像内之各部’計算類似度。繼而’特定類似之程度最 间(於本例中’類似度成為最小之值)之圖像内之位置。例 如’若預先記憶有圖4中例示之基準圖案之圖像及其圖像座 標,則姿勢特定機構14係自平均圖像内,特定與圖4中例示 之基準圖案之圖像之類似度之值成為閾值以下之部位,進 而自其等部位中,特定類似度之值最小之部位。該部位為 於平均圖像内相當於基準圖案之部分。進而,姿勢特定機 構14特定例如所特定之部位之中心像素之圖像座標。即, 164119.doc •28· 201247577 姿勢特定機構14係自平均圖像中特定與圖4中例示之基準 圖案之圖像最類似之部位,特定例如上述中心像素之圖像 座標。姿勢特定機構14針對預先記憶之基準目案之每一圖 像進行該處理。 類似度之計算係利用公知之方法進行即可。例如作為 類似度之例,可列舉SSD(Sum〇fSquaredDifferenee,差值 平方和)或 SAD(Sum of Absolute Difference,絕對差值和)。 SSD係成為類似度算出對象之一對圖像中之對應之像素彼 此之冗度值之差之二次方之合計值。因此,姿勢特定機構 14針對成為類似度算出對象之一對圖像中之每一對應之像 素彼此之組,計算亮度值之差之二次方,進而計算其合計 值,藉此算出SSD即可。又,SAD係成為類似度算出對象之 一對圖像中之對應之像素彼此之亮度值之差之絕對值之合 計值》因此,姿勢特定機構14針對成為類似度算出對象之 一對圖像中之每一對應之像素彼此之組,計算亮度值之差 之絕對值,進而計算其合計值,藉此算出SAD即可。又, 於成為類似度算出對象之圖像為二值圖像之情形時,姿勢 特定機構14亦可針對成為類似度算出對象之一對圖像中之 每一對應之像素彼此之組’計算X〇R(eXclusive 〇R:互斥 邏輯),進而計算其合計值,將其計算結果設為類似度β SSD、SAD及每一像素彼此之組之x〇r之合計值任一者均為 圖像彼此類似之程度越高則值越小之類似度。 再者’於本例令’以使用圖像彼此類似之程度越高則值 越小之類似度之情形為例進行說明,但亦可使用其他類似 164119.doc -29· 201247577The amount of contrast is reduced compared to the previous image. 164119.doc •24· 2012.47577 If the value above the value occurs, up to the image after the fixed time has elapsed, select the image that is not excluded. For example, when the condition is used to calculate the standard deviation of the luminance value as the amount indicating the contrast of light and dark, the preprocessing mechanism 19 is used in a certain image, and the standard deviation of the luminance value is compared with the luminance value of the previous image. When the standard deviation is lower than the fixed value, the image generated from the above time point until the fixed period is excluded from the subsequent processing object, and the remaining image is selected without being excluded. The mechanism 19 generates a preprocessed image based on the selected plurality of images. Further, the fact that the contrast of the light and dark in the image is lowered by the solid value or more means that the contrast is sharply lowered, and it is considered that the raw material powder is flying. In the following description, the case where the preprocessing unit 19 selects an image based on the number of edges in the image will be described as an example. The pre-processing unit 19 uses the selected complex image to determine the luminance values of the respective pixels in the pre-processed image, thereby generating a pre-processed image. In the selected plural image, attention is paid to the corresponding pixel (the same image coordinate pixel), which is the minimum brightness value of the above pixel. Then, the pre-processing unit 19 defines the redundancy value as the brightness value of the corresponding pixel in the pre-processed image, for example, the brightness value of the image coordinate (X1 'yi) read by the processing unit 19 in each selected image. The brightness value in the specific image coordinate (X丨, y,) is the most J value. Then, the preprocessing unit 19 defines the minimum brightness value as the image coordinate of the preprocessed image (the brightness value of χι, _). The pre-mechanism mechanism 19 performs the processing for each pixel. Then, the pre-processing unit 19 memorizes the generated pre-processed image in the image memory mechanism 12. Pre-processing 164119.doc -25· 201247577 Mechanism 19 is fixed period This processing is repeated. Therefore, the preprocessed image generated based on the image obtained by the camera 1込 is sequentially stored in the circular image memory mechanism 12. Further, in the preprocessing, for the plural input from the camera 11a In the image, the image other than the "continuous plural image of the state in which the edge counting result is large" is ignored. Here, the case of using the image obtained by the camera 11a is described as an example. The camera 11b also periodically images the fixed area 9b direction and sequentially inputs the image thereof to the preprocessing mechanism 19. The preprocessing unit 19 also generates an image of the preprocessed image based on the image captured by the camera 11b. And it is memorized in the image memory mechanism 12. It can be said that the continuous plural image of the state in which the counting result of the edge is kept is an image in which the flame or the raw material powder is not photographed. The reason is as follows: In the image of more flame or suspended raw material powder, the batch pile or sidewall becomes unclear, and the number of edges in the image is reduced. Moreover, in the case of shooting a flame, the flame is matched in the image. The brightness value of the part becomes the same value. Therefore, as described above, a plurality of images of the flame or the raw material powder are not imaged, and the minimum brightness value of the corresponding pixels in the image is specified. The brightness value in the image in which the flame or the raw material powder is not photographed can be selected. Since the preprocessed image is specified as the image having the luminance value as described above, even in the case of the camera 11a A raw material powder or flame suspended in the furnace is photographed in one of the obtained images, and a pre-processed image in which the raw material powder or the flame as described above is excluded can be generated. Image of the batch pile. Preprocessor 164119.doc • 26 · 201247577 Structure 19 The action of generating the preprocessed image is equivalent to the pretreatment step β. Again, as described above, the effect on the flame is small, or suspended. In the glass melting furnace with less raw material powder, it is not necessary to perform the pretreatment as described above. In the above case, the image processing device 13 directly records the image obtained by capturing the camera 丨匕, lu into the image memory mechanism. In the case of 丨2, the operation of the posture specifying unit 14 for determining the posture of the camera will be described. Here, taking the case of determining the posture of the camera 11a as an example, the posture determination processing of the camera 1U is also the same. Fig. 8 is a flow chart showing an example of the processing of the posture determination processing of the camera. Further, in the present example, the case where the posture specifying means 14 memorizes the image of the plural reference pattern and the image coordinates thereof will be described as an example. As described above, the pre-processing unit 19 generates a pre-processed image based on the image obtained by the camera 11a for each fixed period (for example, a period of several seconds), and memorizes the image in the image memory mechanism 12. in. The posture specifying unit 14 reads in a plurality of captured images stored in the image memory unit 12 (in this example, the preprocessed image generated based on the image obtained by the camera u & Whether or not the posture of the posture is shifted, wherein the processing period of the posture specifying mechanism 14 is longer than the processing period by the preprocessing mechanism 19, compared to the processing period of the preprocessing unit 19, for example, several seconds. For example, the posture specifying mechanism 14 The processing-specific mechanism 14 may read a captured image of a specific number of sheets that are hidden in the image δ-remembering mechanism 12 when it is determined that the processing start timing is to be performed (according to the camera 11a). The pre-processed image generated by the image obtained by the imaging is 164119.doc -27· 201247577 The predetermined number of sheets is predetermined. The posture specifying unit 14 generates a captured image of a specific number of sheets read in advance (pre- The average image of the image is processed (step S1). Specifically, the gesture-specific mechanism 14 is a camera image for a specific number of sheets read in, for each corresponding pixel. The average value of the luminance values is calculated, and an image in which the average value is set as the luminance value is generated, and the image is assumed to be an average image. In this example, the case of generating an average image is exemplified, but it is also possible to A corresponding pixel calculates an intermediate value of the luminance value, and generates an image (intermediate value image) in which the intermediate value is set as a luminance value. Further, in this example, an average image is generated from the complex image in step S1. In the case of the case, the processing of step S2 and subsequent processing may be performed on one image stored in the image memory unit 12. That is, the processing of step § 亦可 may be omitted. The posture specifying unit 14 generates the step S1. The average image is subjected to pattern matching related to the plural reference pattern previously memorized by the posture specifying mechanism 14 (step S2). » In this example, the higher the degree of similarity of the images, the smaller the value of the calculated similarity. The case will be described as an example. In step S2, the "posture-specific mechanism 14 calculates the similarity between the image of the reference pattern previously memorized and the parts of the average image. Then the degree of the specific similarity is the most (in this case) The position in the image in which the 'similarity becomes the smallest value. For example, 'If the image of the reference pattern illustrated in FIG. 4 and its image coordinates are memorized in advance, the posture specifying mechanism 14 is within the average image. The value of the degree of similarity to the image of the reference pattern exemplified in FIG. 4 is a portion below the threshold value, and the portion having the smallest similarity value from the other portions is the portion which is the reference within the average image. Further, the posture specifying mechanism 14 specifies, for example, the image coordinates of the center pixel of the specific portion. That is, 164119.doc • 28· 201247577 The posture specifying mechanism 14 is specified from the average image and illustrated in FIG. 4 . The most similar portion of the image of the reference pattern specifies, for example, the image coordinates of the center pixel described above. The gesture specific mechanism 14 performs this processing for each image of the reference file that is memorized in advance. The calculation of the degree of similarity can be carried out by a known method. For example, as an example of the degree of similarity, SSD (Sum〇fSquaredDifferenee, Sum of Absolute Difference) or SAD (Sum of Absolute Difference) can be cited. The SSD is a total of the squares of the difference between the ones of the similarity calculations and the redundancy values of the corresponding pixels in the image. Therefore, the posture specifying unit 14 calculates a square of the difference between the luminance values for each of the pixels corresponding to one of the similarity calculation targets, and calculates the total value, thereby calculating the SSD. . Further, the SAD system is a total value of the absolute values of the difference between the luminance values of the corresponding pixels in the image, which is one of the similarity calculation targets. Therefore, the posture specifying unit 14 calculates one of the similarity calculation targets in the image. Each of the corresponding pixels is set to each other, and the absolute value of the difference between the luminance values is calculated, and the total value is calculated, thereby calculating the SAD. Further, when the image to be subjected to the similarity calculation is a binary image, the posture specifying unit 14 may calculate X for the group of each of the pixels corresponding to the one of the similarity calculation objects. 〇R(eXclusive 〇R: mutually exclusive logic), and then calculate the total value, and the calculation result is set to the sum of the similarity degrees β SSD, SAD, and x 〇r of each pixel group. The higher the degree of similarity to each other, the smaller the value is. Further, in the case of the present example, the case where the degree of similarity in which the images are similar to each other is used is described as an example, but other similarities may be used. 164119.doc -29· 201247577
度。例如, (NCC : Norm 類似度。歸- 越是接近1。因此’姿勢特定機構14係於算出 關作為類似度之情形時,特定上述類似度(歸 之值最接近1之部位即可。 ,計算於步驟 繼而’姿勢特定機構14針對每一基準圖案 S2中特定之類似之程度最高之部位(於本例中,為類似度之 值成為最小之部位)之圖像座標與預先記憶之基準圖案之 圖像座標之差(即距離),根據上述距離,判定攝像機丨込之 姿勢是否發生偏移(步驟S3)。姿勢特定機構14將於步驟82 中特定之圖像座標與預先記憶之特徵座標之距離與閾值進 行比較,若座標間之距離為閾值以上,則判定攝像機之姿 勢發生偏移’若座標間之距離未達閾值,則判定攝像機之 姿勢未發生偏移即可。再者,姿勢特定機構14係因預先記 憶有複數之基準圖案,故而針對每一基準圖案而計算座標 間之距離(於步驟S2中特定之圖像座標與預先記憶之特徵 座標之差)。對該複數之距離與閾值進行比較,判定攝像機 之姿勢是否發生偏移之基準並無特別限定。例如,亦可以 針對每一基準圖案進行計算而獲得之複數之座標間距離 中特疋個以上成為閾值以上為條件’判定攝像機之姿勢 發生偏移。或’亦可以所有座標間距離成為閾值以上為條 件’判定攝像機之姿勢發生偏移,此處,例示2個基準,但 亦可依據其他基準而判定攝像機之姿勢是否發生偏移。 164119.doc -30· 201247577 於判定攝像機之姿勢發生偏移之情形(步驟S3中之Yes) 時’姿勢特定機構14將預先記憶之基準圖案之圖像及其圖 像座標之組中之圖像座標置換成於步驟32中特定之圖像座 標’藉此更新記憶之基準圖案之圖像及圖像座標之組中之 圖像座標(步驟S4)。即,姿勢特定機構14係將於平均圖像 内特定為符合基準圖案之部位之部位之圖像座標(於上述 例中,為上述部位之中心像素之圖像座標)作為與上述基準 圖案之圖像成組之圖像座標,更新記憶之圖像座標。藉由 步驟S4之處理,對照攝像機之姿勢之偏移,更新平均圖像 内之基準圖案之座標(圖像座標)。其中,姿勢特定機構14 亦於步驟S5之處理中使用更新前之圖像座標。亦記憶更新 前之圖像座標直至於步驟S5中進行使用為止。 於步驟S4後,姿勢特定機構14使用基準點推斷攝像機na 之姿勢(步驟S5)。於步驟S5中,姿勢特定機構14進行以下 之處理即可。姿勢特定機構14係根據更新前之基準圖案之 圖像座標(預先記憶之基準圖案之圖像座標)與更新後之基 準圖案之圖像座標,計算基準圖案於圖像内以何種程度朝 哪一方向偏移。於存在複數個基準圖案之情形時,例如, 計算每一基準圖案之偏移量之平均值或偏移方向之平均 值’將上述平均值設為基準圖案之偏移量及偏移方向即 可。或,亦可以其他基準,規定更新前後之基準圖案之偏 移量及偏移方向。姿勢特定機構14對照基準圖案之偏移方 向及偏移量’移動預先記憶之基準點之圖像座標。即,對 照更新前後之基準圖案之偏移,更新基準點之圖像座標之 164119.doc •31· 201247577 座標值。繼而’姿勢特定機構14根據實空間中之各基準點 之3維座標算出攝像機Ua之各種姿勢中之各個基準點之圖 像座標。繼而’姿勢特定機構14特定根據各基準點之3維座 標而算出之圖像座標最接近更新後之各基準點之圖像座標 之姿勢,並判定上述姿勢為攝像機lla之姿勢β繼而,結束 指定推斷處理(即步驟S5之處理)。 又’於判定攝像機之姿勢未發生偏移之情形(步驟S3中之degree. For example, (NCC: Norm similarity. The closer to - is closer to 1. Therefore, the posture-specific mechanism 14 is specific to the above-mentioned similarity when calculating the degree of similarity (the portion whose value is closest to 1). The image coordinates and the pre-memorized reference pattern calculated by the posture-specific mechanism 14 for the portion of the highest degree of similarity in each of the reference patterns S2 (in this example, the portion where the value of the similarity is the smallest) is calculated. The difference between the image coordinates (i.e., the distance) determines whether the posture of the camera 发生 is shifted according to the distance (step S3). The posture specific mechanism 14 will specify the image coordinates and the pre-memorized feature coordinates in step 82. The distance is compared with the threshold. If the distance between the coordinates is equal to or greater than the threshold, it is determined that the posture of the camera is shifted. If the distance between the coordinates does not reach the threshold, it is determined that the posture of the camera is not shifted. Since the specific mechanism 14 has a plurality of reference patterns stored in advance, the distance between the coordinates is calculated for each reference pattern (the image specified in step S2) The difference between the target and the pre-memorized feature coordinates. The reference to the complex value is compared with the threshold value, and the criterion for determining whether or not the posture of the camera is shifted is not particularly limited. For example, it may be calculated for each reference pattern. In the case where the number of the coordinates of the plurality of coordinates is equal to or greater than the threshold value, the condition of the camera is determined to be shifted. Alternatively, the posture of the camera may be shifted by the condition that the distance between the coordinates is equal to or greater than the threshold value. It is possible to determine whether or not the posture of the camera is shifted based on the other criteria. 164119.doc -30· 201247577 When the posture of the camera is determined to be shifted (Yes in step S3), the posture specifying mechanism 14 Substituting the image of the pre-memorized reference pattern and the image coordinates in the group of image coordinates with the image coordinates specified in step 32, thereby updating the image of the reference pattern of the memory and the group of image coordinates Image coordinates (step S4). That is, the posture specifying mechanism 14 is a part of the average image that is specified to conform to the reference pattern. The image coordinates of the bit (in the above example, the image coordinates of the central pixel of the above-mentioned portion) are image coordinates grouped with the image of the reference pattern, and the image coordinates of the memory are updated. The processing of step S4 is performed. And updating the coordinates (image coordinates) of the reference pattern in the average image against the offset of the posture of the camera, wherein the posture specifying mechanism 14 also uses the image coordinates before the update in the processing of step S5. The image coordinates are used until the step S5 is used. After the step S4, the posture specifying unit 14 estimates the posture of the camera na using the reference point (step S5). In step S5, the posture specifying unit 14 performs the following processing. The posture specifying mechanism 14 calculates the extent to which the reference pattern is in the image based on the image coordinates of the reference pattern before the update (the image coordinates of the pre-memorized reference pattern) and the image coordinates of the updated reference pattern. Which direction is offset. When there are a plurality of reference patterns, for example, calculating an average value of the offset amounts of each reference pattern or an average value of the offset directions, the average value may be set as the offset amount and the offset direction of the reference pattern. . Alternatively, the offset amount and the offset direction of the reference pattern before and after the update may be specified on other basis. The posture specifying unit 14 moves the image coordinates of the reference point previously memorized in accordance with the shift direction and the offset amount of the reference pattern. That is, the offset of the reference image before and after the update is updated, and the coordinate value of the image coordinates of the reference point is updated to 164119.doc •31·201247577. Then, the posture specifying unit 14 calculates the image coordinates of each of the various postures of the camera Ua based on the three-dimensional coordinates of the reference points in the real space. Then, the posture specifying unit 14 specifies that the image coordinates calculated from the three-dimensional coordinates of the respective reference points are closest to the posture of the image coordinates of the updated reference points, and determines that the posture is the posture β of the camera 11a, and the designation is ended. The inference processing (ie, the processing of step S5). Further, in the case where it is determined that the posture of the camera is not shifted (in step S3)
No)時,姿勢特定機構14係以於步驟S2中特定之平均圖像内 之相當於基準圖案之部位之圖像,更新預先記憶之基準圖 案之圖像(步驟S6)。即,於步驟32中,擷取於平均圖像内 特定為相當於基準圖案之部位之部分之圖像,將上述圖像 作為新基準圖案之圖像而記憶。姿勢特定機構14針對每一 基準圖案進行該處理。藉由該步驟S6之處理,更新姿勢特 疋機構14預先記憶之基準圓案之圖像及其圖像座標之組中 之基準圖案之圖像。 有時玻璃熔融爐内之側壁之狀態緩慢發生變化,而於f 像内相當於基準圖案之部位與姿勢特定機構14記憶之基』 圖案之圖像之類似之程度降低。例如,即便於如圖4所示> 觀察窗之角隅附近之圖案設為基準圖案之情形時,藉由』 料粉緩慢附著於角隅部分’攝像圖像内之基準圖案部分: 圖像有時亦自直角之角隅之圖像緩慢變化成圓形角隅之£ 像。若假設不更新記憶之基準圖案之圖像,則該變化變大 於攝像新圖像時’將無法進行上述圖像與圖4中例示之基』 圖案之圖像之匹配1而’於步驟$5中,根據平均圖像、 164119.doc •32· 201247577 之圖案匹配之結果,更新記情 t籍声灿、隹, 土準圖案之圖像,藉此可 。精又地進仃下一次圖案匹配。例如, 中例示之基準圖案之圖像緩,ρ更 》之圖4 準圖鸯… 為角隅部分為圓形之基 準圖^圖像。其結果,可高精度地進行下―次圖案匹配, 亦可间精度地進行攝像機之姿勢判定。 姿勢特定機構U係對根據攝像機&進行攝像所得之圖 像而生成’且記憶於圖像記憶機構12中之預處理圖像,針 對每一固定之週期,進行步驟sux後之處理即可1樣地, 對根據攝像機llb進行攝像所得之圖像而生成,且記憶於圖 像記憶機構12中之預處理圖像’亦針對每__較之週期, 進行步驟S1以後之處理即可。 又’即便於不進行預處理,而將攝像機lla、llb進行攝像 所得之圖像直接記憶於圖像記憶機構12中之情形時,姿勢 特定機構14亦對攝像機lla進行攝像所得之圖像,針對每一 固定之週期,進行步驟S1以後之處理即可。繼而,同樣地, 對攝像機llb進行攝像所得之圖像,針對每一固定之週期, 進行步驟S1以後之處理即可。 繼而,對作成自正上方觀察時之固定區域中之批料堆之 配置圖像(參照圖7),算出觀察資料之動作進行說明。圖9 係表示該動作之處理經過之例之流程圖。此處,以圖像處 理裝置13對利用攝像機iia所得之攝像圖像(於本例中,為根 據攝像機lla進行攝像所得之圖像而生成之預處理圖像)進 行處理之情形為例進行說明,圖像處理裝置13對利用攝像 機llb所得之攝像圖像(於本例中,根據攝像機llb進行攝像 164119.doc *33· 201247577 所得之圖像而生成之預處理圖像)亦進行相同之處理。 首先,圖像校準機構16按照由新到舊之順序依序讀入複 數張記憶於圖像記憶機構12中之利用攝像機丨la所得之攝 像圖像(於本例中為預處理圖像)。此時讀入之攝像圖像之張 數預先規定即可。繼而’圖像校準機構16自上述各攝像圖 像中’擷取符合實空間中之固定區域93之範圍3 u(參照圖 6)(步驟S10) »所操取之範圍31 a所示之圖像(以下記作擷取 圖像)係以泡為背景之批料堆之圖像。此處,為了方便,以 攝像機11 a之姿勢未發生變化之情形為例進行說明,於攝像 機iia之姿勢發生變化之情形時,圖像校準機構16根據圖像 攝像時之攝像機11 a之姿勢,自攝像圖像中,擷取符合實空 間中之固定區域9a之範圍3ia即可。 再者,於不進行預處理,而分別將攝像機lla、進行攝 像所得之圖像直接記憶於圖像記憶機構丨2中之情形時圖 像校準機構16係於步驟S10中,按照由新到舊之順序依序讀 入複數張由攝像機1 la進行攝像並直接記憶於圖像記憶機 構12中之攝像圖像’自各攝像圖像中將擷取圖像擷取即 可。又,關於由攝像機llb進行攝像並直接記憶於圖像記憶 機構12中之攝像圖像亦相同。關於其他方面,進行預處理 之情形與不進行預處理之情形均相同。步驟sl〇相當於區域 擷取步驟。 繼而,背景圖像作成機構15根據分別自複數張攝像圖像 中擷取之擷取圖像,作成不存在批料堆之情形時之圖像。 即,作成成為批料堆之背景之背景圖像(步驟su)。於步驟 1641 丨 9.doc -34· 201247577 S11中,作成與自最新之攝像圖像中擷取之擷取圖像具有共 通之圖像座標之像素,且上述像素之亮度值表示泡之背景 圖像。步驟S11相當於背景圖像作成步驟。 圖10係表示步驟S11之背景圖像作成處理之處理經過之 例之流程圖。 於背景圖像作成處理中,背景圖像作成機構15選擇自最 新之攝像圖像中擷取之擷取圖像中之各個像素,根據所選 擇之像素及與上述像素對應之其他擷取圖像内之像素之亮 度值,決定於所選擇之像素中表示背景之亮度值。其結果, 獲得不存在批料堆之情形時之背景圖像。以下,參照圖10 對該處理進行說明。再者,此處,以針對每一像素,決定 表示背景之亮度值之情形為例進行說明,背景圖像作成機 構15亦可針對擷取圖像中之每一各個區域,決定表示背景 之亮度值。 责景圖像作成機構15自從最新之攝像圖像中擷取之擷取 圖像之像素中選擇1個像素(步驟S21)。繼而,背景圖像作 成機構15係自於步驟S10(參照圖9)中自其他攝像圖像中擷 取之各擷取圖像中,擷取與所選擇之像素對應之像素(即, 符合固定區域9a内之相同位置之像素)(步驟S22)。 繼而,背景圖像作成機構15以於步驟S21中選擇之像素及 與上述像素對應之其他擷取圖像内之像素(即於步驟s22中 獲得之像素)為對象,針對每一亮度值,計數符合上述亮度 值之像素數(步驟S24)。可以說步驟S24之處理為直方圖作 成處理。 164119.doc •35- 201247577 繼而,背景圖像作成機構15對像素之計數數(度數)變多 之亮度值之範圍内之亮度值之偏差進行評價(步驟s25)。所 謂計數數變多之亮度值之範圍,係指例如計數數成為間值 (對計數數規定之閾值)以上之亮度值連續地持續之範圍。圖 11及圖12係步驟S24之結果所獲得之直方圖。於圖u所示之 例中,像素之計數數變多之亮度值之範圍為^〜匕。於圖12 所示之例中,像素之計數數變多之亮度值之範圍為匕〜、。 作為評價偏差之評價值,使用例如於如上所述之範圍内計 數之像素之亮度值之標準偏差或方差即可。或,將像素之 計數數變多之亮度值之範圍之寬度用作評價值即可。於步 驟S25中,算出如上所述之評價值即可。於算出例示之標準 偏差、方差、或像素之計數數變多之亮度值之範圍之宽度 等作為評價值之情料,評價值越小,則亮度值之偏差越 小。又,亦可將其他指標值用作偏差之評價值。 於步驟S25後,背景圖像作成機構15根據於步驟S25中算 出之《平價值,判疋像素之計數數變多之亮度值之範圍内之 亮度值之偏差是否較大(步驟S26)。於步驟S26中藉由對 預先規定之閾值(對於偏差之評價值之閾值)與評價值進行 比較,狀偏差是否較大即可。例如,於計算亮度值之標 準偏差作為評價值之情形時,若評價值為閾值(對評價值規 疋之閾值)以上,則判定偏差較大,若評價值未達閾值,則 判定偏差較小即可。間值之值根據採用為評價值之指標值 (標準偏差、方差等)預先規定即可。 於判疋亮度值之偏差較小之情形(步驟S26中之N〇)時,背 164119.doc -36 - 201247577 景圖像作成機構15判定計數值變多之亮度值之範圍内之最 頻亮度值(步驟S28)。圖11係亮度值之偏差較小之情形時之 直方圖之例。若以圖11為例,則計數值變多之亮度值之範 圍為,該範圍内之最頻亮度值(像素之計數數成為最 大之亮度值)為S。由此,背景圖像作成機構15於步驟S28中 特定S之值。繼而,將上述s之值決定為於步驟S2i中選擇之 座標之像素中之亮度值。於所選擇之座標中,亮度值之偏 差較小可以指於上述座標中未拍攝有批料堆,而持續拍攝 有背景。由此,於偏差較小之情形時,可如上所述將最頻 亮度值S決定為成為背景之泡之亮度值。再者,亦可如上所 述代替最頻亮度值S’而算出符合計數值變多之亮度值之範 圍ki〜k:2之像素之亮度值之平均值,將上述平均值決定為表 示背景之亮度值。或’亦可將亮度值之範圍ki〜k2之中央值 決定為表示背景之亮度值。 另一方面’於判定亮度值之偏差較大之情形(步驟S26中 之Yes)時,背景圖像作成機構15算出計數值變多之亮度值 之範圍内之符合大於判別基準值之亮度值之各像素之亮度 值之平均值(步驟S27)。圖12係亮度值之偏差較大之情形時 之直方圖之例。若以圖12為例,則計數值變多之亮度值之 範圍為k;3〜k;4。又’判別基準值設為τ。此時,背景圖像作 成機構15計算亮度值符合大於T,且直至h為止之範圍之像 素之亮度值之平均值。繼而’背景圖像作成機構15將上述 平均值決定為於步驟S21中選擇之座標之像素中之亮度 值。於所選擇之座標中’亮度值之偏差較大可以指於上述 164119.doc -37- 201247577 座標中拍攝有批料堆,或拍攝有成為背景之泡。而且,泡 之亮度值大於批料堆之亮度值。由此,可將如上所述符合 大於判別基準值之範圍之像素之亮度值之平均值決定為成 為背景之泡之亮度值。再者,亦可代替如上所述算出平均 值,而判定計數值變多之亮度值之範圍内之大於判別基準 值之範圍(圖12所示之例中為丁〜匕之範圍)中之最頻亮度 值,將上述最頻亮度值決定為所選擇之座標之像素中之亮 度值。或,亦可將T〜k4之範圍中之中央值決定為所選擇之 座標之像素中之亮度值。 再者’判別基準值係用以將偏差較大之範圍(本例中為亮 度值之範圍)分離成2個之閾冑,符合非專利文獻3中記載之 判別分析二值化法_之閾值。因此,將與背景區域與批料 堆區域相關之等級内方差與等級間方差之方差比成為最大 之閾值設為判別基準值T即可。 此處’表示利用判別分析二值化法將亮度值之範圍匕〜 分割成2個等級之情形,亦可利用其他方法,將亮度值之 範圍k3〜k4分割成2個等級。例如,亦可利用模式法或適用2 個常態分佈之方法等將亮度值之範叫〜匕分割成2個等 級。繼而,根據亮度值較高者之等級,與上述同樣地,決 定所選擇之座標之像素中之亮度值即可。 背景圖像作成機構15針對每-像素進行使關ig之流程 圖而說明之上述處理,將於步驟S27或步驟咖中所求得之 亮度值決定為與於步驟S21中所選擇之像素對應之背景圖 像之像素之亮度值。其結果,獲得於自最新之攝像圖像中 164119.doc -38 - 201247577 擷取之擷取圖像中,將批料堆去除之圖像。又該圖像係 以攝像機11 a之視點進行觀察之情形時之背景圖像。 又’背景圖像作成機構1 5亦可針對對擷取圖像進行分割 而獲得之每一各個區域,決定表示背景之亮度值。於此情 形時,於步驟S21中,背景圖像作成機構15自從最新之攝像 圖像中擷取之擷取圖像中選擇丨個區域。區域之規定方法並 無特別限定。繼而’背景圖像作成機構丨5係於步驟S22中, 自從其他攝像圖像中擷取之各掏取圖像中,擷取與所選擇 之區域對應之區域(符合固定區域9a内之相同部分之區 域)。繼而,於步驟S24以後,以屬於在步驟S21中選擇之區 域及與上述區域對應之區域(於步驟S22中獲得之區域)之 各像素為對象,作成直方圖,算出亮度值之偏差之評價值, 根據偏差是否較大’算出亮度值即可(步驟S24〜S28)。背 景圖像作成機構15針對對擷取圖像進行分割而獲得之每一 各個區域進行該處理,將於步驟S27或步驟S28中所求得之 亮度值決定為與於步驟S21中選擇之區域對應之背景圖像 之區域内之各像素之亮度值即可。 於背景圖像作成處理後,轉移至步驟S12(參照圖9)。於 步驟S12中’圖像校準機構16將於背景圖像作成處理(步驟 S11)中所獲得之背景圖像轉換成自正上方觀察固定區域^ 時之圖像(步驟S12)。即,對在步驟S11中所獲得之背景圖 像,進行使視點自攝像機lla之位置變化成固定區域\之正 上方之視點轉換處理,作成自上述視點觀察之情形時之背 景圖像。其結果,於在固定區域9a不存在批料堆之狀態下, 】64119.doc -39· 201247577 獲得自正上方觀察固定區域9a之情形時之圖像。步驟si2相 當於背景圖像轉換步驟。 繼而,圖像校準機構16係於步驟s 10中,將自最新之攝像 圖像中擷取之擷取圖像轉換成自正上方觀察固定區域9&時 之圖像(步驟S13)e即,對自最新之攝像圖像中擷取之擷取 圖像,進行使視點自攝像機lla之位置變化成固定區域&之 正上方之視點轉換處理,轉換成自上述視點觀察之情形時 之圖像。於該轉換後之圖像中,拍攝有批料堆及背景。步 驟S12、S13中之轉換處理為相同之轉換處理。步驟S13相當 於擷取圖像轉換步驟。 再者’於步驟S12、S13中之轉換後之圖像之大小不同之 情形時,圖像校準機構16以使步驟812、S13中之轉換後之 圖像之大小一致之方式進行修正即可。 亦可直接使用於步驟S12、S13中獲得之轉換後之圖像, 執行下述之步驟S14以後之處理。 或’亦可每次檢測最新之攝像圖像時,圖像處理裝置13 執行自步驟sio直至步驟S13為止之處理,圖像校準機構16 分別記憶複數張於步驟S12中所獲得之圖像(自正上方觀察 固定區域9a時之背景圖像)與於步驟S13中所獲得之圖像(自 正上方觀察固定區域9&時之圖像)。繼而,圖像校準機構16 亦可以最新之特定張數部分選擇每次執行步驟S12時所獲 得之圖像’對所選擇之圖像進行合成(例如生成平均圖像), 同樣地’以最新之特定張數部分選擇每次執行步驟S13時所 獲得之圖像,對所選擇之圖像進行合成。繼而,亦可使用 164119.doc 201247577 每次執行步驟S12時所獲得之圖像之合成圖像(自正上方觀 察固定區域9a時之背景圖像)與每次執行步驟sn時所獲得 之圖像之合成圖像(自正上方觀察固定區域9&時之圖像),執 行下述之步驟S14以後之處理。 固體狀態之原料大部分存在於相較液面之下方。因此, 與使用於步驟S12、S13中分別獲得之一張圖像,進行下一 步驟S14以後之處理相比,使用針對每一步驟sn而獲得之 複數張圖像之合成圖像與針對每一步驟S13而獲得之複數 張圖像之合成圖像,進行下一步驟S14以後之處理的方法係 更谷易根據所獲得之觀察資料,把握固體狀態之原料之整 體像。因此,較佳為如上所述,將每次執行步驟Sl2時所獲 得之圖像合成複數張,同樣地,將每次執行步驟S13時所獲 付之圖像合成複數張,使用其等合成圖像,執行步驟S14 以後之處理。 於對每次執行步驟S13時所獲得之複數張之圖像進行合 成之If形時’圖像校準機構16計算例如於複數之圖像中對 應之各像素之亮度值之平均值,將上述平均值設為合成圖 像中之對應像素之亮度值即可。只要針對每一像素進行該 處理’並藉由規定合成圖像之各亮度值而生成合成圖像即 ’亦可代替對應之各像素之亮度值之平均值,對對 應之·各像素之亮度值特定最小值,而將上述亮度值之最小 值設為合成圖像中之對應像素之亮度值。 圖像校準機構16只要藉由於對每次執行步驟S12時所獲 得之複數張圖像進行合成之情形時亦進行相同之處理而生 164119.doc 201247577 成合成圖像即可。 又,於算出批料堆之移動速度作為觀察資料之情形時, 不生成如上所述之合成圖像,而使用於步驟§12、si3中所 獲得之各圖像,進行步驟S14以後之處理即可。又於算出 批料堆之移動速度之情形時,使用攝像機Ua、ub進行攝 像所得之圖像本身,進行步驟S10以後之處理。 繼而,差分運算機構17係於步驟S13之轉換後之圖像與步 驟S12之轉換後之背景圖像之間,算出對應之像素彼此之亮 度值之差(步驟S14)»此處,所謂步驟S13之轉換後之圖像, 既可為於步驟S13中所獲得之丨張圖像,亦可為每次執行步 驟S13時所獲得之複數張圖像之合成圖像。同樣地所謂步 驟S12之轉換後之背景圖像,既可為於步驟中所獲得之ι 張圖像,亦可為每次執行步驟S12時所獲得之複數張圖像之 合成圖像。 於步驟S14中,差分運算機構17自步驟S13之轉換後之圖 像(拍攝有批料堆及背景之圖像)之像素之亮度值,減去步驟 S12之轉換後之背景圖像之像素之亮度值。差分運算機構17 針對每一對應之像素彼此之組進行該減法處理。 圖13係表示步驟s 13之轉換後之圖像之例。於該圖像中, 拍攝有貪景與批料堆1〇。圖14係表示步驟S12之轉換後之背 景圖像之例。圖15係表示對該2個圖像進行步驟S14之處理 所得之結果之圖像之例。如以上說明般,由於泡之亮度亦 多少會有變化,故而於步驟S14之處理後,符合背景之像素 之亮度值未必為〇。 164119.doc -42- 201247577 於步驟SI 4後,差分運算機構π對在步驟S14中獲得之圖 像(參照圖15)進行二值化處理(步驟S15)0即,差分運算機 構17針對圖像内之每一像素,進行將為用於二值化處理而 預先規定之閾值以上之亮度值替換成「丨」,將未達上述閾 值之亮度值替換成「0」之處理。符合背景之像素之亮度值 係因藉由步驟S14之減法處理而變為〇附近之值,故而藉由 二值化處理而變為「〇」。又,符合批料堆1〇之像素之亮度 值因於步驟S14之減法處理中其值不會大幅度減少,故而藉 由二值化處理而變為「1」。其結果,符合背景之像素之亮 度值變為「0」,符合批料堆1〇之像素之亮度值變為Γι」。 將二值化處理後之圖像之例示於圖16中。二值化處理後之 圖像係表示根據自最新之攝像圖像中擷取之擷取圖像而作 成之固定區域9a中之批料堆之位置。再者,該圖像係表示 自固定區域9a之正上方之視點觀察之狀態,不包含批料堆 之高度之資訊。差分運算機構17記憶於步驟S15中生成之圖 像(以下為二值化圖像)。步驟S14、S15相當於背景除外圖 像生成步驟。 於步驟S15後,觀察資料算出機構18使用於步驟S15中生 成之一值化圖像,算出存在於固定區域\内之批料堆之觀 察資料(步驟S16)。其中,於步驟S16中,不僅使用較近生 成之二值化圖像’亦追溯到過去而使用連續之二值化圖像 算出觀察資料。又’此處’對與較區域I相關之二值化 圖像之生成進行了說明,圖像處理裝置13根據利用攝像機 lib所得之攝像圖像,亦生成與固定區域、相關之二值化圖 I64119.doc -43· 201247577 像。觀察資料算出機構18亦可根據固定區域9b各自之一 值化圖像算出觀察資料。步驟S16相當於觀察資料算出步 驟。 以下,表示於步驟S16_算出之觀察資料之例。作為觀察 資料之例,可列舉固定區域9a、9b各自之内外比。圖17係表 不將固定區域9a、9b平分成側壁6側之區域與玻璃熔融爐之 中央側之區域所得之區域的說明圖。對於與圖丨所示之要素 相同之要素,標註與圖1相同之符號並省略說明。區域51 ' 52係將固定區域9a平分成側壁6側之區域與中央側之區域 所得之區域,區域51為側壁6側之區域,區域52為中央側之 區域。同樣地,區域41、42係將固定區域9b平分成側壁6側 之區域與中央側之區域所得之區域,區域41為側壁6側之區 域,區域42為中央側之區域。觀察資料算出機構18計算表 不區域51内之批料堆之佔有率與區域52内之批料堆之佔有 率之比之評價值作為與固定區域9a相關之内外比即可。 又,同樣地,si*算表示區域41内之批料堆之佔有率與區域 42内之批料堆之佔有率之比之評價值作為與固定區域^相 關之内外比即可。 例如,於將側壁6側之區域(即區域51或區域41)中之批料 堆之佔有率設為Q,將中央側之區域(即區域52或區域42)中 之批料堆之佔有率設為汉之情形時,觀察資料算出機構18 亦可計算以下之式(1 )所示之評價值作為内外比。其中,Q、 R係由百分率表示,分別為〇〜1〇〇之範圍之值。 内外比=(R—Q)/(R+q+α) 式(1) 164119.doc • 44 - 201247577 於式(1)中α為常數’例如,亦可設為α=ι〇〇^於此情形 時,内外比成為-0.5〜0.5之範圍之值。觀察資料算出機構 18對固定區域9a、9b分別算出内外比即可。 若固體狀態之原料過度靠近於側壁6側,則有時原料在未 溶解之狀態下自玻璃熔融爐中流出,於上述之情形時,玻 璃之品質降低。藉由内外比,可確認固體狀態之原料是否 過度靠近於側壁6側。於判斷固體狀態之原料過度靠近於侧 壁6側之情形時,以批料堆向中央靠近之方式,操作玻璃熔 融爐即可。 又,觀察資料算出機構18亦可算出固定區域9a、9b各自中 之批料堆之佔有率。 又,觀察資料算出機構18亦可算出固定區域9a、9b各自中 之批料堆之頂端位置(例如批料堆之頂端位置之座標)。 又,因批料堆之狀態或火焰之蔓延等,有於二值化圖像 中未拍攝有批料堆之頂端位置之情形。於此情形時,觀察 資料算出機構18沿與熔解之原料之前進方向垂直之方向分 割固定區域9a,算出各分割區域中之批料堆之面積。繼而, 亦可將自上游側之分割區域朝向下游侧之分割區域方向之 批料堆之面積之變化視為線形變化,算出批料堆之面積變 為0之位置,並將上述位置判定為批料堆之頂端位置。關於 固定區域9b亦相同。 若批料堆之頂端位置過度延伸至下游側,則產生在未熔 解之狀態下流出之可能性。於觀察資料算出機構18算出之 批料堆之頂端位置過度延伸至下游側之情形時,以批料堆 164119.doc •45· 201247577 之頂端位置恢復至上游側之方式操作玻璃熔融爐即可。 又,觀察資料算出機構18亦可算出自上游側觀察時右側 之固定區域\中之觀察資料之值與自上游側觀察時左侧之 固疋區域9b中之觀察資料之值之差作為觀察資料。例如, 觀察資料算出機構18亦可算出固定區域^中之批料堆之佔 有率與固定區域9b中之批料堆之佔有率之差。又,觀察資 料算出機構18亦可算出固定區域\中之批料堆之頂端位置 與固定區域9b中之批料堆之頂端位置之差。以下,將固定 區域9a、9b中之觀察資料之值之差記作左右差。藉由亦算出 該左右差作為觀察資料之W,可確認自上游側觀察時右侧 與左側中固體原料之狀態是否無偏向。例如,可確認自上 游側觀察時僅於右側與左側之任一者中進行熔解,而於另 一者中轉延遲等狀況,可以根據上述狀況,操作玻璃炫 融爐之方式進行判斷。 例如,於藉由在步驟S16中算出之左右差,判斷固定區域 9a、9b之任—者中之原料之料延遲之情料,進行增加對 於原料之轉延遲之—方之“固定區域 之燃料投入量(即,加強燃燒器之火力)等操作即可』 再者於上述例中,對算出批料堆之佔有率或與頂端位 置相關之左右差之情形進行了說明,但觀察資料算出機構 18亦可算出與其他觀察資料相關之左右差。 又’批料堆之佔有率、 較近之1張二值化圖像而算 化圖像之合成圖像而算出 頂端位置及其等左右差既可根據 出’亦可根據較近之複數之二值 。再者,如以上說明般,就把握 164119.doc •46- 201247577 固體狀態之原料之整體像之觀點而言,較佳為對在每一步 驟S12中獲得之各圖像進行合成,且對在每一步驟中獲 得之各圖像進行合成,使用其等合成圖像,進行步驟si4 以後之處理’生成二值化圖像。 又,觀察資料算出機構18亦可根據連續之複數之二值化 圖像中之同一批料堆之位置與攝像機之攝像間隔,算出批 料堆整體之移動速度。由於批料堆整體之移動緩慢,故而 於連續之複數之二值化圖像中,同一批料堆之位置之變化 較少。由此,觀察資料算出機構18判定於連續之複數之二 值化圖像中,位置座標最近之批料堆彼此為同一批料堆即 可。繼*,根據同一批料堆之座標之變化,#出上述批料 堆之移動距離,根據上述移動距離與攝像間隔算出批料堆 整體之移動速度即可。於本例中,將一個批料堆之移動速 度視為批料堆整體之移動速度H於算出批料堆之速 度作為觀察資料之情形時,使用攝像機m、m進行攝像 所得之圖像本身,進行步驟S1G以後之處理。進而,使用於 步驟S12中所獲得之m圖像與於步驟S13中所獲得之i張圖 像’進行步驟S14以後之處理。 又’觀察資料算出機構18亦可根據連續之複數之二值化 圖像中之同-批料堆之位置’算出批料堆之移動方向。 觀察資料算出機構18亦可根據連續之複數之二值化 圖像’算出批料堆之減少率。你丨k IT me 减^羊例如’觀察資料算出機構18 亦可於連續之各二值化圖像中, 列疋冋一批料堆,於各二 值化圖像中,算出批料堆 曲積或長度之減少 164119.doc -47· 201247577 於算出長度之減少率時,既可根據沿著原料 =出減少率,或亦可根據沿著與原料 = 之方向之長度算出減少率。 万向垂直 :二觀察資料算出機構18較佳為於算 移動速度、批料堆之移動隹整體之 田嗤嬙把 批枓堆之減少率等時,使 用連續之複數之二值化圖像,亦可使料連 值化圖像。 吸数之一 該批料堆之減少率係於與批料堆之高度之減少率 :間存在相關’可藉由批料堆之減少率判斷批料堆之高 :。右批料堆過高’則溶解需要花費時間,而頂端位置延 伸0 進而,觀察資料算出機構18亦可根據二值化圖像,算妇 各個批料堆之方向(批料堆延伸之方向)。如上所述之方向^ 預先規定成為基準之方向,藉由與上述基準方向形成之, 度表示即可》 又’觀察資料算出機構18亦可根據二值化圖像,算出各 個批料堆之大小。 又,觀察資料算出機構18亦可根據二值化圖像與於步驟 ⑴中獲得之圖像’計算評價批料堆中之氣體之喷出狀態之 ”平價值。若氣體自批料堆中喷出,則於圖像内觀察到陷沒 於批料堆之表面之孔而看上去粗糙。由此,觀察資料算出 機構18亦可使用二值化圖像判定於步驟su中所獲得之圖 像中相當於批料堆之區域,計算上述區域中之亮度值之標 準偏差而將上述標準偏差設為氣體之喷出狀態之評價值。 164119.doc -48- 201247577 又’因氣體之喷出而陷沒之部分係觀察為較黑之區域β 由此,觀察資料算出機構18亦可使用二值化圖像判定於步 驟S13中所獲得之圖像中相當於批料堆之區域,計數上述區 域内之黑色像素之總數,將其計數結果設為氣體之喷出狀 態之評價值。 於非專利文獻1或專利文獻1中,記載有對批料堆之佔有 率或批料堆之頂端位置(最下游位置)進行評價,於本發明 中’並不限定於其等,藉由測定内外比、左右差、批料堆 之速度或移動方向、批料堆之減少率、各個批料堆之方向 或大小、批料堆中之氣體之噴出狀態之評價值等多種觀察 資料’可穩定地進行批料堆之定量評價β X,根據上述結 果,可藉由確切地運轉玻璃熔融爐而製造高品質之玻璃。 又,根據本發明,姿勢特定機構14對攝像圖像(更具體而 言為攝像圖像之平均圖像)進行基準圖案之圖案匹配,根據 攝像圖像内之基準圖案之圖像座標,判定攝像機之姿勢之 偏移之有無,於判定發生偏移之情形時,使用姿勢之偏移 量’特定攝像機之姿勢(位置及方向)。繼而,圖像校準機構 16根據攝像機之姿勢,自攝像圖像_擷取符合實空間中之 固定區域9a、9b之範圍。進而,背景圖像作成機構15根據上 述擷取圖像作成背景圖像,圖像校準機構16對擷取圖像及 背景圖像進行使視點自攝像機之位置變化成固定區域之正 上方之視點轉換處理’差分運算機構17計算兩者之亮度值 之差分。因此,即便於清掃時等攝像機之姿勢發生變化, 亦可良好地持續玻璃熔融爐内之固定區域之觀察。 164119.doc • 49- 201247577 ••預處理’預處理機構19係自從攝像機輸入之複 之像t,選擇保持邊緣之計數結果較多之狀態之連續 之複數之圖像。繼而,預處理機構19係於所選擇之複數之 圖像中’著眼於對應之像素,特定上述像素中成為最小之 亮度值’將上述亮度值規定為預處理圖像中之對應像素之 亮度值。預處理機構19針對每一對應像素進行該處理。根 據攝像機進行攝像所得H亦有拍攝有於爐内懸浮之 原料粉,或拍攝有火焰,而背景或批料堆變得不清楚之情 形i~藉由如上所述進行預處理,可作成火焰或原料粉等 干擾之影響較少之圖像。繼而,藉由使用如上所述之圖像, 進行步驟S1G以後之處理(參照㈣,可獲得干擾之影響較 〆之良好之背景圖像或僅表示批料堆之圖像亦良好之圖 像,而可正確地監視固定區域中之批料堆之狀態。 由此,根據本實施形態,可持續玻璃熔融爐内之固定區 域之觀察,而良好地監視上述固定區域中之批料堆之狀 態。再者,如以上說明般,於監視原料粉或火焰之影響較 少之玻璃熔融爐之情形時,亦可不進行預處理。於上述之 情形時,亦可使用攝像機對爐内進行攝像而生成之圆像本 身’進行步驟S10以後之處理(參照圖9)。 又’於本實施形態中,姿勢特定機構14對攝像圖像進行 複數之基準圖案之圖案匹配,特定攝像機之姿勢。如此般, 藉由使用複數之基準圖案,攝像機之姿勢偏移判定之可靠 性增加。 繼而’對第1實施形態之變形例進行說明。於上述第i實 164119.doc .50· 201247577 施形態中,表示對自背景圖像及最新之攝像圖像中擷取之 擷取圖像,分別進行轉換處理(步驟S12、S13,參照圖9)後, 進^計算ϋ分之處理(步驟S14,參照圖9)n亦^先 進行計算像素彼此之差分之處理後,進行轉換處理。圖Μ 係表示如上所述之第1#施形態之變形例中之破璃溶融爐 内監視系統之構成例的方塊圖。圖18所示之各機構係與圖2 所示之各機構相同之機構,以與圖2相同之符號表示。然 而,於本變形例中’由於各種圖像之流程之一部分與上述 第1實施形態不同,故而表示各種圖像之流程之箭頭與圖2 不同。又’圖19係表示如上所述之第!實施形態之變形例中 之直至觀察資料算出為止之處理經過之例的流程圖。關於 與第1實施形態中已說明之處理相同之處理標註與圖9相同 之符號’並省略說明。 於本變形例中,於步驟810、S11後,差分運算機構⑽ 於自最新之攝像圖像中擷取之榻取圖像與於步驟川中作 成之背景圖像之間,算出對應之像素彼此之亮度值之差(步 驟S3!)。此時,差分運算機構17自從最新之攝像圖像中擷 取之擷取圖像(拍攝有批料堆及背景之圖像)之像素之亮度 值中’減去背景圖像之像素之亮度值。差分運算機構”針 對每一對應之像素彼此之組進行該減法處理。其結果,獲 付自攝像機之視點進行觀察所得之固定區域之圖像,且將 背景去除之圖像。然而,於上述之減法結果中,符合背景 之像素之亮度值未必變為〇。 因此,差分運算機構17係於步驟S31後,對於步驟S31中 164II9.doc •51 · 201247577 所獲得之圖像進行二值化處理(步驟S32) ^其結果,獲得自 攝像機之視點進行觀察所得之固定區域之圖像,且符合背 景之像素之亮度值為「〇」’符合批料堆ίο之像素之亮度值 變為「1」之二值化圖像。步驟S3丨、S32相當於背景除外圖 像生成步驟。 於步驟S32後’圖像校準機構16對在步驟S32中所生成之 二值化圖像’進行使視點自攝像機之位置變化成固定區域 之正上方之視點轉換處理(步驟S33)。其結果,獲得與於以 上說明之步驟Sl5(參照圖9)中獲得之二值化圖像相同之二 值化圖像。步驟S33相當於背景除外圖像轉換步驟。 於步驟S32後’觀察資ί斗算出機構18使用步驟S32中之轉 換處理後之二值化圖像,算出存在於固定區域内之批料堆 之觀察資料(步驟S16)。該處理係與以上說明之步驟S16之 處理相同。 [實施形態2] 圖20係表示本發明之第2實施形態之玻璃熔融爐内監視 系統之構成例的方塊圖。與第1實施形態相同之構成要素係 標註與圖2相同之符號,並省略說明。第2實施形態之玻璃 炫1融爐内監視系統包含攝像機lla、攝像機llb及圖像處理裝 置13a〇圖像處理裝置除包含預處理機構19、圖像記憶機 構12、姿勢特定機構14、背景圖像作成機構15、圖像校準 機構16、差分運算機構17及觀察資料算出機構18以外,亦 包含觀察資料解析機構61與熔融爐控制機構62。又,圖像 處·理裝置13a亦可為對圖18所示之玻璃熔融爐内監視系統 164119.doc -52- 201247577 之圖像處理裝置追加觀察資料艇 構62所得之構成。 解析機構61及溶融爐控制機 觀察資料解析機構61對藉由觀察資料算出機構18算出值 之多種觀察資料與玻璃炫融爐之多種運轉參數之相關之程 度進行判定。換言之,觀察資料解析機構61導出玻璃溶融 爐之多種運轉參數對藉由觀察資料算出機㈣算出值之多 種觀察f料帶來之影響之程度。作為觀察資料之例,可列 舉固定區域9a、9b各自中之批料堆之佔有率、批料堆之頂端 位置及其等觀察資料之左右差、固定區域%中之内外 比、批料堆之移動速度、批料堆之減少率等,但觀察資料 不阳定於該等X,作為運轉參數,可列舉燃燒器燃料 之燃燒條件(例如燃燒量等)、原料之投入條件(例如投入量 等)、批料.玻璃屬t匕等’但運轉參數亦不限定於該等。 觀察資料解析機構61係藉由例如主成分分析及多變量解 析(例如複回歸分析),判定觀察資料與運轉參數之相關之程 度例如,觀察資料解析機構61進行主成分分析而求出主 成分,利用上述主成分進行多變量解析。繼而,觀察資料 解析機構61係藉由利用於上述過程中使用之係數,導出各 參數之影響度。所謂參數之影響度,具體而言,係指運轉 參數對觀察資料帶來之影響之程度。觀察資料解析機構6ι 導出參數之影響度之處理相當於影響度導出步驟。 圖21係表示計算運轉參數對丨個觀察資料(此處設為觀察 資料A)之影響度所得之結果之例的圖表。於圖2丨中,表示 觀察資料A與作為運轉參數之投入條件A(設為原料之投入 164U9.doc -53- 201247577 量)、投入條件B、燃燒參數A〜D之相關。圖21之縱轴為各 運轉參數之影響度。燃燒參數A〜D為各場所之燃燒器中之 燃燒量。若運轉參數之影響度之值為正,則於與觀察資料 之間有正相關,若運轉參數之影響度之值為負,則於與觀 察資料之間有負相關。又’影響度之值之絕對值越大,則 表示運轉參數與觀察資料之相關之程度越強。 例如,若根據圖21所示之結果,使投入條件a(原料之投 入量)增加,則意味著觀察資料A之值亦增加。又,使若燃 燒參數A增加,則意味著觀察資料a之值減少。 熔融爐控制機構62係參照藉由觀察資料算出機構18而算 出之觀察資料,於上述觀察資料達到應變更玻璃熔融爐之 運轉狀況之值後,變更於與上述觀察資料之間具有相關性 之運轉參數。此處,所謂於與觀察資料之間具有相關性之 運轉參數,係指例如對觀察資料之影響度之絕對值成為預 先規定之值以上之運轉參數。例如,於觀察資料之值超過 上限值,過度變高之情形時,使於與上述觀察資料之間具 有正相關之運轉參數之值減少或使於與上述觀察資料之間 具有負相關之運轉參數之值增加。又,例如,於觀察資料 之值未達下限值,過度變低之情形時,使於與上述觀察資 料之間具有正相關之運轉參數之值增加或使於與上述觀察 資料之間具有負相關之運轉參數之值減少。作為具體例 於判定在作為觀察資料之批料堆之佔有率與作為運轉參數 之爐内溫度之間有負相關,且批料堆之佔有率超過上限值 之情形時,熔融爐控制機構62係以使爐内溫度上升之方式 1641 丨 9.doc • 54· 201247577 操作破璃熔融爐即可。即,使燃燒器之火力上升即可。如 此般熔融爐控制機構62變更運轉參數之處理相當於熔融爐 控制步驟。 又,熔融爐控制機構62亦可於觀察資料之值超過上限值 或未達下限值時輸出警報。 再者,玻璃熔融爐之運轉參數之變更亦可由操作人員進 行°於此情形時,亦可不包含熔融爐控制機構62。又,於 此情形時,操作人員參照藉由觀察資料算出機構18而算出 之觀察資料與藉由觀察資料解析機構61而算出之觀察資料 與運轉參數之間之影響度,判斷如何變更哪一運轉參數即 可。 根據本實施形態,由於觀察資料解析機構61算出表示運 轉參數相對於觀察資料之相關之程度之影響度,故而可使 根據監視之批料堆之狀態,調節玻璃㈣爐之哪—運轉參 數即可明確化。 進而,藉由設置熔融爐控制機構62,可不通過操作人員, 而自動地將玻璃熔融爐控制為確切之狀態。 ;述說月巾表示觀察資料解析機構61算出運轉參數 對觀察資料之影響度之情形。除此以外,於獲得表示原料 狀態之品質之品質資料(例如泡個數等)之情形時,觀察資料 解析機構61亦可算出表示觀察f料或運轉參數相對於品質 ^之相關之程度之影響度。該影響度亦可藉由例如主成 =析及多變量解析而進行。再者,泡個數越多,則意味 者爐之狀態不佳。 164119.doc -55- 201247577 圖22係表示計算觀察資料a、B及作為運轉參數之溫度a. 〜D對作為1個品質資料之泡個數之影響度所得之結果的圖 表。觀察資料A、B係觀察資料算出機構18藉由根據攝像圖 像而生成之二值化圖像而算出之資料。溫度A〜D係藉由對 玻璃熔融爐之各場所之溫度進行計測而獲得之值。於圖22 所示之例中,若影響度之值為正,則亦於觀察資料或溫度 與品質資料之間有正相關,若影響度之值為負,則亦於觀 察資料或溫度與品質資料之間有負相關。又,影響度之值 之絕對值越大’則表示相關之程度越大。 例如,根據圖22所示之結果,可知觀察資料a、B之值越 大,泡個數越是增加(品質變差)β又,可知溫度A之值越低, 泡個數越是增加。 再者,即便於某條件下,判定於某觀察資料與品質資料 之間存在相關,有時亦於其他條件下,判定於其他觀察資 料與上述品質資料之間亦存在相關。圖23係表示觀察資料 與品質資料之相關失去或重新出現之狀況之變化的圖表。 圖23所示之左側之縱軸表示觀察資料之值。右側之縱軸表 不品質資料(此處為泡個數)之值。橫軸表示時間之經過。於 圖23所示之例中,直至計測期間之中途為止,於觀察資料a 與品質資料之間確認到相關,於後半部分中,上述相關失 去。又,直至計測期間之中途為止,於觀察資料B與品質資 料之間不存在相關,於後半部分中,確認到觀察資料B與品 質資料之相關。 因此’觀察資料解析機構61較佳為反覆算出觀察資料與 164119.doc •56· 201247577 品質資料之間之影響度。 再者’於第2實施形態中,表示根據觀察資料解析機構61 算出之影響度,判定與觀察資料相關之運轉參數,根據觀 察資料變更上述運轉參數之情形《於操作人員參照二值化 圖像’可判斷應操作哪一運轉參數之情形時,操作人員亦 可參照二值化圖像,增減運轉參數。例如,於根據二值化 圖像,判斷自上游壁觀察而右側之批料堆之熔解延遲之情 形時’操作人員亦可自上游壁觀察而使右側之燃燒器之火 力上升。 又’於上述各實施形態中’亦可將攝像機1“配置於自正 上方攝像固定區域9a之位置,將攝像機iib配置於自正上方 攝像固定區域9b之位置。於此情形時,特徵性之物(例如侧 壁、燃燒器等)亦包含於攝像範圍中,亦攝像有基準圖案或 基準點。如此般,於攝像機1 la配置於自正上方攝像固定區 域9a之位置,攝像機llb配置於自正上方攝像固定區域%之 位置之情形時,亦可不進行使視點變化成固定區域\或固 定區域9b之正上方之視點轉換處理。即,亦可不進行步驟 S12 ' S13(參照圖9)之視點轉換處理。又,亦可於作為實施 形態之變形例而表示之處理經過(參照圖19)中,不進行步驟 S33之視點轉換處理。 [實施形態3] 繼而’作為本發明之第3實施形態,對玻璃物品之製造方 法進行說明。於本發明之玻璃物品之製造方法中,應用第^ 實施形態中已說明之玻璃熔融爐内監視方法。進而,亦可 164119.doc •57· 201247577 將第2實施形態中已說明之觀察資料與運轉參數之相關之 程度之判定、及運轉參數之變更處理應用於本發明之玻璃 物品之製造方法。圖24係表示本實施形態之玻璃物品之製 造方法中使用之玻璃物品之製造線之一例的模式圖。再 者’於圖24中,省略攝像機lla、llb及圖像處理裝置13之圖 示’於玻璃熔融爐1之附近配置有攝像機lla、llb。又,亦 配置有圖像處理裝置13。然而,圖像處理裝置13之配置位 置並無限定。又’亦可配置第2實施形態中已說明之圖像處 理裝置13a。 於玻璃物品之製造線中,設置有玻璃熔融爐1與澄清槽 30。再者,澄清槽30之種類並無限定。澄清槽3〇亦可為使 槽之内部為減壓狀態而去除泡之減壓型澄清槽。或,澄清 槽30亦可為使槽之内部為高溫而去除泡之高溫型澄清槽。 玻璃熔融爐1(參照圖24及圖1)使玻璃原料熔解,使其變 化成熔融玻璃71。於圖24中,省略批料堆之圖示。澄清槽 30去除熔融玻璃71中產生之泡。將泡去除之熔融玻璃係轉 移至成形步驟、緩冷步驟。 圖25係表示本實施形態之玻璃物品之製造方法之例的流 程圖。首先,將玻璃原料投入至玻璃熔融爐1。破璃溶融爐 1包含燃燒器5(參照圖1),將玻璃熔融爐1之内部維持於高 溫。繼而’藉由於玻璃溶融爐1中加熱玻璃原料,製造溶融 玻璃71(步驟S91、玻璃熔融步驟)。 於步驟S91中,攝像機lla、llb對玻璃熔融爐1之内部進行 攝像,圖像處理裝置13對其結果所獲得之圖像進行與第i 164119.doc • 58 · 201247577 實施形態相同之處理。即,進行步驟S51〜S54(參照圖5)、 步驟S1〜S6(參照圖8)、步驟S10〜S16(參照圖9或圖19) '步 驟S21〜S28(參照圖10)等處理。藉由該處理,獲得觀察資 料,可良好地監視玻璃熔融爐1之内部。又,於第2實施形 態中已說明之圖像處理裝置13a亦可與第2實施形態同樣 地’判定觀察資料與玻璃熔融爐1之運轉參數之相關之程 度,變更玻璃熔融爐1之運轉參數。 於步驟S91中所製造之熔融玻璃71係流至澄清槽3卜於該 熔融玻璃71中存在泡,於溶融玻璃71之表面產生泡層(省略 圖示)。於澄清槽30之内部,去除熔融玻璃71之泡(步驟 S92、澄清步驟)。 於步驟S92後’成形將泡去除之熔融玻璃(步驟S93、成形 步驟)。於成形步驟中,例如,藉由浮式法成形熔融玻璃即 可。具體而言’藉由使將泡去除之熔融玻璃71浮於熔融錫 (未圖不)上’並使其向搬送方向前進而製成連續之板狀之玻 璃帶此時,為了成形特定板厚之玻璃帶,而按壓於玻璃 帶之兩側部分旋轉之輥’向寬度方向(與搬送方向成直角之 方向)外側拉伸玻璃帶。 繼而,使於步驟S93中成形之玻璃帶緩冷(步驟S94、緩冷 步驟)。於緩冷步驟中’自熔融錫中拉出玻璃帶,於緩冷爐 (未圖不)之内部使玻璃帶緩慢冷卻。即便於搬送至緩冷爐之 外部後’亦進-步使玻璃帶緩冷至常溫左右為止。 /於緩冷步驟後’視需要對在緩冷步驟中固化之玻璃帶進 行加工(步驟S95、加工步驟)。作為步驟奶中之加工例, 164119.doc •59- 201247577 可列舉例如切斷或研磨。然而,並不限定於切斷或研磨, 亦可進行其他加工處理。 根據本實施形態之玻璃物品之製造方法,可良好地持續 玻璃溶融爐内之固定區域之觀察,並且製造玻璃物品。尤 其’若圖像處理裝置13a與第2實施形態同樣地,判定觀察 資料與玻璃溶融爐1之運轉參數之相關之程度,變更玻璃炫 融爐1之運轉參數,則可以與爐内之觀察結果相對應之確切 之運轉參數運轉玻璃熔融爐1,而製造玻璃物品》 以上已參照特定之實施態樣詳細地說明了本出願,業者 應明白可在不脫離本發明之精神與範圍之狀態下施加多種 變更或修正。 本出願係基於2011年5月6日申請之日本專利出願(曰本 專利特願2011 -103601)者,其内容係以參照之形式寫入於 此。 [產業上之可利用性] 本發明較佳地應用於監視玻璃熔融爐内之批料堆之玻璃 熔融爐内監視系統。 【圖式簡單說明】 圖1係表示應用本發明之玻璃炫融爐内監視系統之玻璃 馆·融爐之例的俯視圖。 圖2係表示本發明之第1實施形態之玻璃溶融爐内監視系 統之構成例的方塊圖》 圖3係表示利用攝像機1 la所得之攝像圖像之例之說明 圖。 1641l9.doc • 60· 201247577 圖4(a)、4(b)係表示基準圖案之圖像之例及使用基準圖案 之匹配之例的說明圖。 圖5係表示姿勢特定機構14進行之姿勢推斷動作之例之 流程圖。 圖6係抽出利用攝像機1 1 &所得之攝像圖像中符合溶解之 原料之液面之範圍的模式圖。 圖7係表示以使視點變化成固定區域93之正上方之方式 進行轉換所得之轉換結果之例的說明圖。 圖8係表示攝像機之姿勢判斷處理之處理經過之例之流 程圖。 圖9係表示直至觀察資料導出為止之處理經過之例之流 程圖。 圖10係表示背景圖像作成處理(步驟Sii)之處理經過之 例之流程圖。 圖11係步驟S24之結果所獲得之直方圖。 圖12係步驟S24之結果所獲得之直方圖。 圖13係表示步驟S13之轉換後之圖像之例的說明圖。 圖14係表示步驟S12之轉換後之背景圖像之例的說明圖。 圖15係表示進行步驟S14之處理所得之結果之圖像之例 的說明圖。 圖16係表示二值化處理後之圖像之例之說明圖。 圖17係表示將固定區域9 a、9 b平分成側壁6側之區域與玻 璃熔融爐之中央側區域所得之區域的說明圖。 圖18係表示第1實施形態之變形例中之玻璃熔融爐内監 164119.doc -61 ^ 201247577 視系統之構成例的方塊圖。 圖19係表示第!實施形態之變形例中之直至觀察資料導 出為止之處理經過之例的流程圖。 圖2 0係表示本發明之第2實施形態之玻璃溶融爐内監視 系統之構成例的方塊圖。 圖21係表示計算運轉參數對丨個觀察資料之影響度所得 之結果之例的圖表。 圖22係表示計算觀察資料a、B及溫度A〜D對1個品質資 料之影響度所得之結果的圖表》 圖23係表示觀察資料與品質資料之相關失去或重新出現 之狀況之變化的圖表。 圖24係表示第3實施形態之玻璃物品之製造方法中使用 之玻璃物品之製造線之一例的模式圖。 圖25係表示第3實施形態之玻璃物品之製造方法之例的 流程圖。 【主要元件符號說明】 2 3a 3b 4 5 6 7 玻璃熔融爐 觀察窗 投入口 投入口 排出口 燃燒器 側壁 上游壁 164119.doc -62- 201247577 8 下游壁 9a 固定區域 9b 固定區域 10 批料堆 11a 攝像機 lib 攝像機 12 圖像記憶機構 13 圖像處理裝置 13a 圖像處理裝置 14 姿勢特定機構 15 背景圖像作成機構 16 圖像校準機構 17 差分運算機構 18 觀察資料算出機構 19 預處理機構 21a 點 21b 點 21〇 點 21c 點 21e 點 25 液面 30 澄清槽 31a 範圍 41 區域 164119.doc .63· 201247577 42 區域 51 區域 52 區域 61 觀察資料解析機構 62 熔融爐控制機構 71 熔融玻璃 81 攝像圖像内之部分 SI 步驟 S2 步驟 S3 步驟 S4 步驟 S5 步驟 S6 步驟 S10 步驟 Sll 步驟 S12 步驟 S13 步驟 S14 步驟 S15 步驟 S16 步驟 S21 步驟 S22 步驟 S24 步驟 S25 步驟 164119.doc 201247577 526 527 528 531 532 533 551 552 553 554 591 592 593 594 595 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 步驟 164119.doc -65In the case of No), the posture specifying unit 14 updates the image of the reference pattern previously stored in the image of the portion corresponding to the reference pattern in the average image specified in the step S2 (step S6). That is, in step 32, an image of a portion corresponding to a portion corresponding to the reference pattern in the average image is captured, and the image is stored as an image of the new reference pattern. The posture specifying mechanism 14 performs this processing for each reference pattern. By the processing of this step S6, the image of the reference circle and the image of the reference pattern in the group of image coordinates which are previously memorized by the posture specializing means 14 are updated. The state of the side wall in the glass melting furnace may change slowly, and the degree of similarity between the portion corresponding to the reference pattern in the f image and the image of the base pattern stored in the posture specifying mechanism 14 may be lowered. For example, even if the pattern near the corner 观察 of the observation window is set as the reference pattern as shown in FIG. 4, the reference powder portion in the image is imaged by the powder slowly attached to the corner portion: image Sometimes the image from the right angle of the corner is slowly changed into a rounded corner image. If it is assumed that the image of the reference pattern of the memory is not updated, the change becomes larger than the matching of the image of the image of the image illustrated in FIG. 4 when the new image is captured, and the result is changed in step $5. , according to the average image, 164119. Doc •32· 201247577 The result of the pattern matching, update the t-sound t-can, 隹, the image of the quasi-pattern, which can be used. Finely enter the next pattern match. For example, the image of the reference pattern exemplified in the figure is slow, and Fig. 4 is a quasi-image of the figure 鸯. As a result, the next-order pattern matching can be performed with high precision, and the posture determination of the camera can be performed with high precision. The posture-specific mechanism U generates a pre-processed image stored in the image memory mechanism 12 based on the image captured by the camera & and performs the processing after the step sux for each fixed cycle. For example, the pre-processed image 'generated by the image captured by the camera 11b and stored in the image memory unit 12 may be subjected to the processing of step S1 and subsequent steps for each period. Further, even when the image obtained by capturing the cameras 11a and 11b is directly stored in the image memory unit 12 without performing preprocessing, the posture specifying unit 14 also images the image obtained by the camera 11a. For each fixed period, the processing after step S1 may be performed. Then, similarly, the image obtained by imaging the camera 11b may be subjected to the processing in and after step S1 for each fixed period. Next, the operation of calculating the observation data by referring to the arrangement image of the batch pile in the fixed area when viewed from the upper side (see Fig. 7) will be described. Fig. 9 is a flow chart showing an example of the processing of the action. Here, the case where the image processing device 13 processes the captured image obtained by the camera iia (in this example, the preprocessed image generated based on the image obtained by the camera 11a) is described as an example. The image processing device 13 pairs the captured image obtained by the camera 11b (in this example, the camera llb is imaged 164119. Doc *33· 201247577 The pre-processed image generated by the obtained image) is also processed in the same way. First, the image calibration unit 16 sequentially reads a plurality of image images (in this example, preprocessed images) obtained by the camera 丨la in the image memory mechanism 12 in the order from newest to oldest. The number of images to be read at this time may be predetermined. Then, the 'image calibration mechanism 16 draws a range from the above-mentioned respective captured images' that matches the range 3 u of the fixed area 93 in the real space (see FIG. 6) (step S10) » the range 31 a taken The image (hereinafter referred to as the captured image) is an image of the batch pile with the bubble as the background. Here, for the sake of convenience, the case where the posture of the camera 11 a does not change will be described as an example. When the posture of the camera iia changes, the image calibration mechanism 16 is based on the posture of the camera 11 a at the time of image capturing. From the captured image, the range 3ia conforming to the fixed area 9a in the real space can be extracted. Furthermore, when the camera 11a and the image obtained by the imaging are directly memorized in the image memory mechanism 丨2 without pre-processing, the image calibration mechanism 16 is in step S10, according to the new to the old The sequence is sequentially read in a plurality of captured images captured by the camera 1 la and directly memorized in the image memory mechanism 12, and the captured images are captured from the respective captured images. Further, the captured image captured by the camera 11b and directly memorized in the image memory unit 12 is also the same. Regarding other aspects, the case of performing pre-processing is the same as the case of not performing pre-processing. Step sl 〇 is equivalent to the area capture step. Then, the background image creating means 15 creates an image in the case where there is no batch pile based on the captured images taken from the plurality of captured images. That is, a background image that becomes the background of the batch pile is created (step su). In step 1641 丨 9. Doc -34· 201247577 In S11, a pixel having a common image coordinate with the captured image captured from the latest captured image is created, and the luminance value of the pixel represents the background image of the bubble. Step S11 corresponds to a background image creation step. Fig. 10 is a flow chart showing an example of the processing of the background image creation processing of step S11. In the background image creation processing, the background image creation unit 15 selects each pixel in the captured image captured from the latest captured image, and selects the image according to the selected pixel and the corresponding pixel. The brightness value of the pixel within the pixel is determined by the brightness value representing the background in the selected pixel. As a result, a background image in the case where there is no batch heap is obtained. This processing will be described below with reference to Fig. 10 . Here, the case where the luminance value indicating the background is determined for each pixel will be described as an example, and the background image creating unit 15 may determine the brightness of the background for each of the regions in the captured image. value. The responsible image creation means 15 selects one pixel from among the pixels of the captured image extracted from the latest captured image (step S21). Then, the background image creating unit 15 extracts pixels corresponding to the selected pixels from the captured images captured from the other captured images in step S10 (refer to FIG. 9) (ie, conforms to the fixed The pixel of the same position in the area 9a) (step S22). Then, the background image creating unit 15 counts the pixels selected in the step S21 and the pixels in the other captured images corresponding to the pixels (ie, the pixels obtained in step s22), and counts for each brightness value. The number of pixels satisfying the above luminance value (step S24). It can be said that the processing of step S24 is a histogram creation processing. 164119. Doc 35-201247577 Then, the background image creating means 15 evaluates the deviation of the luminance values in the range of the luminance value in which the number of counts (degrees) of the pixels is increased (step s25). The range of the luminance value in which the number of counts is increased is, for example, a range in which the number of counts is equal to or greater than the luminance value (the threshold value defined for the number of counts). 11 and 12 are histograms obtained as a result of step S24. In the example shown in Fig. u, the range of luminance values in which the number of pixels counts is increased is ^~匕. In the example shown in FIG. 12, the range of luminance values in which the number of counts of pixels is increased is 匕~. As the evaluation value of the evaluation deviation, for example, the standard deviation or the variance of the luminance values of the pixels counted in the range as described above may be used. Alternatively, the width of the range of the luminance value in which the number of counts of the pixels is increased may be used as the evaluation value. In step S25, the evaluation value as described above may be calculated. The case where the standard deviation, the variance, or the width of the range in which the number of counts of the pixels is increased is calculated as the evaluation value, and the smaller the evaluation value, the smaller the variation in the luminance value. Further, other index values may be used as the evaluation value of the deviation. After the step S25, the background image creating means 15 determines whether or not the deviation of the luminance value within the range of the luminance value in which the number of counts of the pixels is increased is large according to the "flat value" calculated in the step S25 (step S26). In step S26, it is sufficient to compare the predetermined threshold value (threshold value of the evaluation value for the deviation) with the evaluation value, and whether the shape deviation is large. For example, when calculating the standard deviation of the luminance value as the evaluation value, if the evaluation value is equal to or greater than the threshold value (the threshold value for the evaluation value), the determination deviation is large, and if the evaluation value does not reach the threshold value, the determination deviation is small. Just fine. The value of the inter-value may be predetermined based on the index value (standard deviation, variance, etc.) used as the evaluation value. In the case where the deviation of the luminance value is small (N〇 in step S26), the back is 164119. Doc -36 - 201247577 The scene image creating means 15 determines the maximum frequency luminance value within the range of the luminance value in which the count value is increased (step S28). Fig. 11 is an example of a histogram in the case where the deviation of the luminance values is small. Taking Fig. 11 as an example, the range of the luminance value in which the count value is increased is such that the maximum frequency luminance value (the number of counts of pixels becomes the largest luminance value) in the range is S. Thereby, the background image creating mechanism 15 specifies the value of S in step S28. Then, the value of s above is determined as the luminance value in the pixel of the coordinates selected in step S2i. In the selected coordinates, the smaller deviation of the brightness values may mean that the batch is not photographed in the above coordinates, and the continuous shooting has a background. Therefore, when the deviation is small, the maximum-frequency luminance value S can be determined as the luminance value of the bubble which becomes the background as described above. Further, instead of the most frequent luminance value S', the average value of the luminance values of the pixels of the range ki~k:2 corresponding to the luminance value in which the count value is increased may be calculated as described above, and the average value may be determined to represent the background. Brightness value. Alternatively, the central value of the range of luminance values ki to k2 may be determined to represent the luminance value of the background. On the other hand, when it is determined that the deviation of the luminance value is large (Yes in step S26), the background image creating unit 15 calculates the luminance value in the range of the luminance value in which the count value is increased to be larger than the discrimination reference value. The average of the luminance values of the respective pixels (step S27). Fig. 12 is an example of a histogram in the case where the deviation of the luminance values is large. Taking Fig. 12 as an example, the range of the luminance value in which the count value is increased is k; 3 to k; Further, the discrimination reference value is set to τ. At this time, the background image forming unit 15 calculates an average value of the luminance values of the pixels whose luminance values match the range of more than T and up to h. Then, the background image creating means 15 determines the above average value as the luminance value in the pixel of the coordinates selected in step S21. The deviation of the brightness value in the selected coordinates may be referred to above as 164119. Doc -37- 201247577 The coordinates of the batch are taken in the coordinates, or the bubbles are the background. Moreover, the brightness value of the bubble is greater than the brightness value of the batch. Thereby, the average value of the luminance values of the pixels which satisfy the range larger than the discrimination reference value as described above can be determined as the luminance value of the bubble which becomes the background. Further, instead of calculating the average value as described above, it is also possible to determine the range among the range of the luminance value in which the count value is increased, which is larger than the range of the discrimination reference value (in the example shown in FIG. 12, the range of the range of 丁 to 匕) The frequency brightness value determines the above-mentioned maximum frequency brightness value as the brightness value in the pixel of the selected coordinate. Alternatively, the median value in the range of T to k4 may be determined as the luminance value in the pixel of the selected coordinate. Further, the 'determination reference value' is used to separate the range in which the deviation is large (the range of the luminance value in this example) into two thresholds, which is in accordance with the threshold value of the discriminant analysis binarization method described in Non-Patent Document 3. . Therefore, the threshold value at which the variance ratio between the intra-level variance and the inter-level variance associated with the background region and the batch pile region is maximized is set as the discrimination reference value T. Here, 'the case where the range of the luminance value 匕~ is divided into two levels by the discriminant analysis binarization method, and the range of the luminance values k3 to k4 can be divided into two levels by other methods. For example, the mode of the luminance value can be divided into two levels by the mode method or the method of applying two normal distributions. Then, in accordance with the level of the higher luminance value, the luminance value in the pixel of the selected coordinate may be determined in the same manner as described above. The background image creation means 15 performs the above-described processing for the process of turning off the ig for each pixel, and the luminance value obtained in step S27 or step coffee is determined to correspond to the pixel selected in step S21. The brightness value of the pixels of the background image. The result was obtained from the latest camera image 164119. Doc -38 - 201247577 Extract the image from the batch heap in the captured image. Further, the image is a background image when viewed from the viewpoint of the camera 11a. Further, the background image creating means 15 may determine the brightness value indicating the background for each of the respective regions obtained by dividing the captured image. In this case, in step S21, the background image creating means 15 selects one of the captured images extracted from the latest captured image. The method of specifying the area is not particularly limited. Then, the 'background image creation mechanism 丨5 is in step S22, and the regions corresponding to the selected region are captured from the captured images captured from the other captured images (the same portion in the fixed region 9a is matched) Area). Then, after step S24, each pixel belonging to the region selected in step S21 and the region corresponding to the region (the region obtained in step S22) is subjected to a histogram, and an evaluation value of the deviation of the luminance value is calculated. It is sufficient to calculate the brightness value based on whether the deviation is large (steps S24 to S28). The background image creating unit 15 performs the processing for each of the respective regions obtained by dividing the captured image, and the luminance value obtained in step S27 or step S28 is determined to correspond to the region selected in step S21. The brightness value of each pixel in the area of the background image may be. After the background image creation processing, the process proceeds to step S12 (see FIG. 9). In step S12, the image calibration mechanism 16 converts the background image obtained in the background image creation processing (step S11) into an image in which the fixed area is observed from directly above (step S12). In other words, the background image obtained in step S11 is subjected to viewpoint conversion processing for changing the position of the viewpoint from the position of the camera 11a to the upper side of the fixed area \ to create a background image when viewed from the viewpoint. As a result, in the state where the batch pile is not present in the fixed area 9a, 】 64119. Doc -39· 201247577 Obtain an image when the fixed area 9a is observed from the upper side. Step si2 corresponds to the background image conversion step. Then, the image calibration mechanism 16 converts the captured image captured from the latest captured image into an image when the fixed region 9& is viewed from directly above (step S13) e, that is, in step s10, The image captured from the latest captured image is subjected to viewpoint conversion processing for changing the position of the viewpoint from the position of the camera 11a to the upper side of the fixed area & and converted into an image when viewed from the viewpoint described above. . In the converted image, a batch and a background are taken. The conversion processing in steps S12, S13 is the same conversion processing. Step S13 is equivalent to capturing the image conversion step. Further, when the sizes of the images after the conversion in steps S12 and S13 are different, the image calibration unit 16 may correct the sizes of the images after the conversion in steps 812 and S13. It is also possible to directly use the converted image obtained in steps S12 and S13, and perform the processing of step S14 and subsequent steps described below. Or 'may also detect the latest captured image each time, the image processing device 13 performs the processing from step sio to step S13, and the image calibration mechanism 16 memorizes the plurality of images obtained in step S12, respectively. The background image when the fixed area 9a is observed directly above) and the image obtained in the step S13 (the image when the fixed area 9& is observed from the upper side). Then, the image calibration mechanism 16 can also select the image obtained each time the step S12 is executed to synthesize the selected image (for example, generate an average image), and the same as the latest one. The specific number of sheets portion selects an image obtained each time the step S13 is performed, and the selected image is synthesized. Then, you can also use 164119. Doc 201247577 A composite image of the image obtained each time the step S12 is executed (the background image when the fixed area 9a is viewed from the upper side) and a composite image of the image obtained each time the step sn is executed (self-positive The image of the fixed area 9& is viewed from above, and the processing of step S14 and subsequent steps described below is performed. Most of the raw materials in the solid state exist below the liquid level. Therefore, compared with the one image obtained in each of steps S12 and S13, the composite image of the plurality of images obtained for each step sn is used for each of the subsequent steps S14 and subsequent processes. The method of performing the processing of the next step S14 after the composite image of the plurality of images obtained in step S13 is based on the obtained observation data, and grasps the overall image of the raw material in a solid state. Therefore, it is preferable to synthesize a plurality of images obtained each time the step S12 is performed as described above, and similarly, combine the images obtained at the time of executing the step S13 into a plurality of images, and use the composite image thereof. For example, the processing after step S14 is performed. The image calibration mechanism 16 calculates an average value of the luminance values of the corresponding pixels in the complex image, for example, when the image of the plurality of images obtained at each step S13 is synthesized. The value is set to the brightness value of the corresponding pixel in the composite image. As long as the processing is performed for each pixel 'and the composite image is generated by specifying the luminance values of the composite image, that is, 'the average value of the luminance values of the corresponding pixels can be replaced, and the luminance value of the corresponding pixel is A specific minimum value is set, and the minimum value of the above luminance values is set as the luminance value of the corresponding pixel in the composite image. The image calibration mechanism 16 generates 164119 by performing the same processing for the case of synthesizing the plurality of images obtained each time the step S12 is performed. Doc 201247577 can be combined into an image. Further, when the moving speed of the batch pile is calculated as the observation data, the composite image as described above is not generated, and the respective images obtained in steps § 12 and si3 are used, and the processing in and after step S14 is performed. can. Further, when calculating the moving speed of the batch pile, the image itself obtained by the use of the cameras Ua and ub is subjected to the processing in and after step S10. Then, the difference calculation unit 17 calculates the difference between the luminance values of the corresponding pixels between the image after the conversion in step S13 and the background image after the conversion in step S12 (step S14). Here, step S13 is performed. The converted image may be the 丨 image obtained in step S13 or the composite image of the plurality of images obtained each time step S13 is performed. Similarly, the background image after the conversion of the step S12 may be either the ι image obtained in the step or the composite image of the plurality of images obtained each time the step S12 is performed. In step S14, the difference calculation mechanism 17 subtracts the pixel value of the converted background image of step S12 from the brightness value of the image of the image after the conversion of step S13 (the image of the batch pile and the image is captured). Brightness value. The difference arithmetic unit 17 performs the subtraction processing for each of the corresponding pixels. Fig. 13 is a diagram showing an example of the image after the conversion of the step s13. In this image, there is a greedy scene and a batch of shots. Fig. 14 is a view showing an example of a background image after the conversion of step S12. Fig. 15 is a view showing an example of an image obtained by performing the processing of step S14 on the two images. As described above, since the brightness of the bubble changes somewhat, the luminance value of the pixel corresponding to the background is not necessarily 〇 after the processing of step S14. 164119. Doc -42 - 201247577 After step SI 4, the difference calculation means π binarizes the image (see Fig. 15) obtained in step S14 (step S15), that is, the difference arithmetic means 17 is for the image. For each pixel, a process of replacing the luminance value equal to or greater than a predetermined threshold value for the binarization processing with "丨" and replacing the luminance value not reaching the threshold value with "0" is performed. The luminance value of the pixel corresponding to the background is changed to the value near 〇 by the subtraction processing of step S14, and thus becomes "〇" by the binarization processing. Further, the luminance value of the pixel corresponding to the stack of the batch is not significantly reduced by the subtraction processing in the step S14, and therefore becomes "1" by the binarization processing. As a result, the luminance value of the pixel corresponding to the background becomes "0", and the luminance value of the pixel corresponding to the batch of the batch becomes Γι". An example of the binarized image is shown in FIG. The binarized image represents the position of the batch pile in the fixed area 9a which is created based on the captured image taken from the latest captured image. Further, the image indicates the state of viewing from the viewpoint directly above the fixed area 9a, and does not include information on the height of the batch pile. The difference calculation unit 17 memorizes the image generated in step S15 (hereinafter, the binarized image). Steps S14 and S15 correspond to the background exclusion image generation step. After the step S15, the observation data calculation unit 18 generates the observation data of the batch pile existing in the fixed area\ using the one-valued image in the step S15 (step S16). Here, in step S16, not only the near-produced binarized image ‘ but also the past is used, and the continuous binarized image is used to calculate the observation data. Further, 'herein' describes the generation of the binarized image associated with the region I, and the image processing device 13 also generates a binarized map associated with the fixed region based on the captured image obtained by the camera lib. I64119. Doc -43· 201247577 like. The observation data calculation means 18 can also calculate the observation data based on the one-valued image of each of the fixed areas 9b. Step S16 corresponds to the observation data calculation step. Hereinafter, an example of the observation data calculated in step S16_ will be described. Examples of the observation data include the internal and external ratios of the fixed regions 9a and 9b. Fig. 17 is an explanatory view showing a region obtained by not dividing the fixing regions 9a and 9b into a region on the side of the side wall 6 and a region on the center side of the glass melting furnace. The same elements as those shown in the drawings are denoted by the same reference numerals as in FIG. 1 and their description will be omitted. The region 51' 52 is formed by dividing the fixed region 9a into a region on the side of the side wall 6 and a region on the center side, the region 51 being the region on the side of the side wall 6, and the region 52 being the region on the center side. Similarly, the regions 41 and 42 are formed by dividing the fixed region 9b into a region obtained by the region on the side of the side wall 6 and the region on the center side, the region 41 being the region on the side of the side wall 6, and the region 42 being the region on the center side. The observation data calculation means 18 calculates the evaluation value of the ratio of the occupancy rate of the batch pile in the area 51 to the occupancy rate of the batch pile in the area 52 as the ratio of the inside to the outside of the fixed area 9a. Further, in the same manner, the evaluation value of the ratio of the occupancy rate of the batch pile in the area 41 to the occupancy rate of the batch pile in the area 42 can be regarded as the ratio of the inside and the outside of the fixed area. For example, the occupancy rate of the batch pile in the area on the side of the side wall 6 (ie, the area 51 or the area 41) is set to Q, and the occupation rate of the batch pile in the area on the center side (ie, the area 52 or the area 42) is set. When it is set to the case of Han, the observation data calculation means 18 can also calculate the evaluation value shown by the following formula (1) as an internal-external ratio. Among them, Q and R are expressed as a percentage, which are values in the range of 〇~1〇〇. Internal and external ratio = (R - Q) / (R + q + α) Formula (1) 164119. Doc • 44 - 201247577 α is a constant in equation (1). For example, it can also be set to α=ι〇〇^. In this case, the internal and external ratio becomes -0. 5~0. The value of the range of 5. The observation data calculation unit 18 may calculate the internal and external ratios for each of the fixed regions 9a and 9b. When the raw material in a solid state is excessively close to the side wall 6, the raw material may flow out from the glass melting furnace in an undissolved state, and in the above case, the quality of the glass is lowered. By the ratio of the inside to the outside, it can be confirmed whether or not the material in the solid state is excessively close to the side wall 6 side. When it is judged that the material of the solid state is excessively close to the side of the side wall 6, the glass melting furnace can be operated in such a manner that the batch pile is approached toward the center. Further, the observation data calculation means 18 can also calculate the occupation ratio of the batch pile in each of the fixed areas 9a and 9b. Further, the observation data calculation means 18 can also calculate the position of the top end of the batch pile in each of the fixed areas 9a, 9b (e.g., the coordinates of the top end position of the batch pile). Further, due to the state of the batch pile or the spread of the flame, there is a case where the top position of the batch pile is not photographed in the binarized image. In this case, the observation data calculating means 18 divides the fixed area 9a in a direction perpendicular to the advance direction of the molten material, and calculates the area of the batch pile in each divided area. Then, the change of the area of the batch pile in the direction from the divided area on the upstream side toward the divided area on the downstream side can be regarded as a linear change, and the area where the batch pile area becomes 0 is calculated, and the above position is determined as a batch. The top position of the pile. The same applies to the fixed area 9b. If the top end position of the batch pile is excessively extended to the downstream side, there is a possibility that it will flow out in an unmelted state. When the position of the top end of the batch pile calculated by the observation data calculation unit 18 is excessively extended to the downstream side, the batch pile 164119 is used. Doc •45· 201247577 The position of the top end can be restored to the upstream side. Further, the observation data calculation means 18 can also calculate the difference between the value of the observation data in the fixed area on the right side when viewed from the upstream side and the value of the observation data in the solid area 9b on the left side when viewed from the upstream side as observation data. . For example, the observation data calculation unit 18 can also calculate the difference between the occupancy rate of the batch pile in the fixed area and the occupancy rate of the batch pile in the fixed area 9b. Further, the observation data calculation means 18 can also calculate the difference between the top end position of the batch pile in the fixed area \ and the top end position of the batch pile in the fixed area 9b. Hereinafter, the difference between the values of the observation data in the fixed areas 9a and 9b is recorded as the left-right difference. By calculating the left and right difference as the observation data W, it is confirmed whether or not the state of the solid material in the right side and the left side is not biased when viewed from the upstream side. For example, it can be confirmed that the melting is performed only in either of the right side and the left side when viewed from the upstream side, and the other one is delayed in the middle or the like, and the glass smelting furnace can be operated in accordance with the above situation. For example, by determining the left and right difference calculated in step S16, it is determined that the material of the fixed area 9a, 9b is delayed, and the fuel of the fixed area is added to increase the delay of the material. In the above example, the amount of input (that is, the heating power of the burner is increased) can be described. In the above example, the case where the occupancy rate of the batch pile or the difference between the top and bottom positions is calculated is described, but the observation data calculation mechanism is used. 18 can also calculate the left and right differences associated with other observations. Also calculate the top position and its difference between the left and right sides of the composite image of the image of the batch pile and the near binarized image. Can be based on the 'can also be based on the value of the nearest plural. Again, as explained above, grasp 164119. Doc • 46- 201247577 From the viewpoint of the overall image of the raw material in a solid state, it is preferable to synthesize each image obtained in each step S12, and synthesize each image obtained in each step, Using the composite image, the processing of step si4 and subsequent steps is performed to generate a binarized image. Further, the observation data calculation means 18 can calculate the moving speed of the entire batch pile based on the position of the same batch pile in the continuous plural number of binarized images and the imaging interval of the camera. Due to the slow movement of the batch pile as a whole, the position of the same batch of stocks changes less in successive binary images. Thus, the observation data calculation unit 18 determines that the batch piles closest to the position coordinates are the same batch pile in the binary image of the continuous plurality. Following *, according to the change of the coordinates of the same batch of stock, the moving distance of the above batch pile is calculated, and the moving speed of the batch pile is calculated according to the above moving distance and the imaging interval. In this example, the moving speed of a batch pile is regarded as the moving speed H of the whole batch pile. When the speed of the batch pile is calculated as the observation data, the image obtained by the camera m and m is used to capture the image itself. The processing after step S1G is performed. Further, the processing of step S14 and subsequent steps is performed using the m image obtained in step S12 and the i image obtained in step S13. Further, the observation data calculation means 18 can calculate the moving direction of the batch pile based on the position of the same-batch pile in the continuous plural number of binarized images. The observation data calculation means 18 can also calculate the reduction rate of the batch pile based on the continuous plural number of binarized images. You 丨k IT me minus ^ sheep, for example, 'observation data calculation mechanism 18 can also list a batch of stocks in successive binarized images, and calculate the batch pile in each binarized image. The product or length is reduced by 164119. Doc -47· 201247577 When calculating the reduction rate of the length, the reduction rate can be calculated based on the reduction rate along the raw material = or the length along the direction from the raw material =. Universal vertical: The second observation data calculation means 18 preferably uses a continuous complex binary image when calculating the moving speed, the movement of the batch pile, the overall reduction rate of the stack, and the like. It is also possible to bin the image. One of the suction numbers The reduction rate of the batch is based on the reduction rate from the height of the batch: there is a correlation between the batch piles and the reduction of the batch pile: If the right batch pile is too high, it takes time to dissolve, and the top position extends 0. Further, the observation data calculation mechanism 18 can also calculate the direction of each batch pile according to the binarized image (the direction in which the batch pile extends). . The direction as described above is defined in advance as the direction of the reference, and the degree is expressed by the reference direction. The observation data calculation unit 18 can also calculate the size of each batch pile based on the binarized image. . Further, the observation data calculation means 18 can also calculate the flat value of the ejection state of the gas in the batch pile based on the binarized image and the image obtained in the step (1). If the gas is sprayed from the batch pile When it is observed, the hole which is trapped in the surface of the batch pile is observed in the image and looks rough. Thus, the observation data calculation means 18 can also determine the image obtained in the step su using the binarized image. In the area corresponding to the batch pile, the standard deviation of the brightness values in the above area is calculated, and the standard deviation is set as the evaluation value of the gas discharge state. 164119. Doc -48- 201247577 Further, the portion which is trapped by the gas ejection is observed as the darker region β. Thus, the observation data calculation means 18 can also determine the map obtained in the step S13 using the binarized image. In the area corresponding to the batch pile in the image, the total number of black pixels in the above area is counted, and the count result is set as an evaluation value of the gas discharge state. In Non-Patent Document 1 or Patent Document 1, it is described that the occupancy rate of the batch pile or the tip position (the most downstream position) of the batch pile is evaluated, and in the present invention, it is not limited to the above, and is determined by Internal and external ratio, left and right difference, speed or moving direction of batch pile, reduction rate of batch pile, direction or size of each batch pile, evaluation value of gas discharge state in batch pile, etc. The quantitative evaluation of β X of the batch pile is carried out, and according to the above results, high-quality glass can be produced by operating the glass melting furnace in an exact manner. Further, according to the present invention, the posture specifying unit 14 performs pattern matching of the reference pattern on the captured image (more specifically, the average image of the captured image), and determines the camera based on the image coordinates of the reference pattern in the captured image. The presence or absence of the deviation of the posture is used to determine the posture (position and direction) of the specific camera when the offset is determined. Then, the image calibration mechanism 16 captures the range of the fixed areas 9a, 9b in the real space from the captured image_ according to the posture of the camera. Further, the background image creating unit 15 creates a background image based on the captured image, and the image calibration unit 16 performs a viewpoint conversion of the captured image and the background image so that the viewpoint is changed from the position of the camera to directly above the fixed area. The 'differential calculation mechanism 17' calculates the difference between the luminance values of the two. Therefore, even when the posture of the camera changes during cleaning, the observation of the fixed area in the glass melting furnace can be favorably continued. 164119. Doc • 49- 201247577 ••Preprocessing The pre-processing mechanism 19 selects a continuous multiplicity of images in a state in which the counting result of the edge is large, since the complex image t input from the camera. Then, the pre-processing mechanism 19 defines the brightness value as the brightness value of the corresponding pixel in the pre-processed image by focusing on the corresponding pixel and specifying the minimum brightness value among the pixels in the selected plurality of images. . The pre-processing mechanism 19 performs this processing for each corresponding pixel. According to the camera, the H is also photographed with the raw material powder suspended in the furnace, or the flame is photographed, and the background or batch pile becomes unclear. i~ It can be made into a flame by pretreatment as described above. An image with less influence on interference such as raw material powder. Then, by using the image as described above, the processing after step S1G is performed (refer to (4), and a good background image with a better influence of interference or an image indicating that the image of the batch pile is good is also obtained. Further, the state of the batch pile in the fixed area can be accurately monitored. Thus, according to the present embodiment, the state of the batch pile in the fixed area can be satisfactorily monitored by observing the fixed area in the glass melting furnace. Further, as described above, in the case of monitoring the glass melting furnace having less influence of the raw material powder or the flame, the pretreatment may not be performed. In the above case, the camera may be used to image the inside of the furnace. In the present embodiment, the posture specifying unit 14 performs pattern matching of the plurality of reference patterns on the captured image to specify the posture of the camera. Thus, the borrowing is performed. The reliability of the posture deviation determination of the camera is increased by using the reference pattern of the plural. Then, a modification of the first embodiment will be described. 64119. Doc . 50·201247577 In the embodiment, the captured image is extracted from the background image and the latest captured image, and the conversion process is performed (steps S12 and S13, see FIG. 9), and then the calculation is performed. Processing (step S14, refer to FIG. 9) n also performs processing for calculating the difference between the pixels, and then performs conversion processing. Fig. 方块 is a block diagram showing a configuration example of a glass-melting furnace monitoring system in a modification of the first embodiment. The mechanism shown in Fig. 18 is the same as the mechanism shown in Fig. 2, and is denoted by the same reference numeral as Fig. 2. However, in the present modification, since one of the flows of various images is different from the first embodiment described above, the arrows indicating the flow of various images are different from those of Fig. 2 . Further, Fig. 19 shows the above as described above! In the modified example of the embodiment, a flow chart of an example of the processing until the observation of the observation data is performed. The same processes as those described in the first embodiment are denoted by the same reference numerals as those in Fig. 9 and the description thereof will be omitted. In the present modification, after the steps 810 and S11, the difference computing means (10) calculates the corresponding pixels between the image taken from the latest captured image and the background image created in the step Chuan. The difference in luminance values (step S3!). At this time, the difference computing mechanism 17 subtracts the luminance value of the pixel of the background image from the luminance value of the pixel captured from the latest captured image (the image of the batch and the image is captured) . The difference operation means performs the subtraction processing for each of the corresponding pixels. As a result, an image of the fixed area observed from the viewpoint of the camera is taken, and the background is removed. However, in the above In the subtraction result, the luminance value of the pixel corresponding to the background does not necessarily become 〇. Therefore, the difference operation mechanism 17 is after step S31, and is 164II9 in step S31. Doc •51 · 201247577 The image obtained is binarized (step S32). ^ As a result, an image of the fixed area observed from the viewpoint of the camera is obtained, and the brightness value of the pixel matching the background is "〇" 'Benary image that matches the brightness value of the pixel of the batch heap ίο to "1". Steps S3 and S32 correspond to the background exclusion image generation step. After step S32, the image correcting means 16 performs a viewpoint conversion process of changing the position of the viewpoint from the position of the camera to the immediately above the fixed area for the binarized image generated in step S32 (step S33). As a result, a binarized image identical to the binarized image obtained in the above-described step S15 (refer to Fig. 9) is obtained. Step S33 corresponds to the background exclusion image conversion step. After the step S32, the observation processing unit 18 calculates the observation data of the batch pile existing in the fixed area using the binarized image after the conversion processing in step S32 (step S16). This processing is the same as the processing of step S16 described above. [Embodiment 2] FIG. 20 is a block diagram showing a configuration example of a glass-melting furnace monitoring system according to a second embodiment of the present invention. The same components as those in the first embodiment are denoted by the same reference numerals as those in Fig. 2, and the description thereof will be omitted. The glass-light-melting-in-furnace monitoring system of the second embodiment includes a camera 11a, a camera 11b, and an image processing device 13a. The image processing device includes a pre-processing unit 19, an image memory unit 12, a posture-specific mechanism 14, and a background image. The observation data analysis unit 61 and the melting furnace control unit 62 are included in addition to the image forming unit 15, the image calibration unit 16, the difference calculation unit 17, and the observation data calculation unit 18. Further, the image processing device 13a may be the glass melting furnace monitoring system 164119 shown in Fig. 18. The image processing device of doc-52-201247577 adds the structure obtained by the observation data boat 62. The analysis unit 61 and the melting furnace controller The observation data analysis unit 61 determines the degree of correlation between the plurality of observation data calculated by the observation data calculation unit 18 and the plurality of operation parameters of the glass furnace. In other words, the observation data analysis unit 61 derives the extent to which the various operating parameters of the glass melting furnace affect the various observations of the calculated value by the observation data calculation unit (4). As an example of the observation data, the occupancy rate of the batch pile in each of the fixed areas 9a and 9b, the position of the top end of the batch pile, and the left and right difference of the observation data, the ratio of the inside to the outside of the fixed area, and the batch pile can be cited. The moving speed, the reduction rate of the batch pile, etc., but the observation data is not determined by the X. As the operating parameters, the combustion conditions of the burner fuel (for example, the amount of combustion, etc.) and the input conditions of the raw materials (for example, the amount of input, etc.) ), batch materials. The glass is t匕, etc., but the operating parameters are not limited to these. The observation data analysis unit 61 determines the degree of correlation between the observation data and the operation parameters by, for example, principal component analysis and multivariate analysis (for example, complex regression analysis). For example, the observation data analysis unit 61 performs principal component analysis to determine the principal component. Multivariate analysis is performed using the above principal components. Then, the observation data analysis unit 61 derives the influence degree of each parameter by using the coefficient used in the above process. The degree of influence of the parameters, in particular, refers to the extent to which the operating parameters affect the observed data. The process of observing the degree of influence of the parameter analysis means 6 ι is equivalent to the influence degree deriving step. Fig. 21 is a graph showing an example of the results of calculating the influence of the operating parameters on one observation data (herein, observation data A). In Fig. 2, the observation data A and the input condition A as the operation parameter are shown (the input of the raw material is 164U9. Doc -53- 201247577 quantity), input condition B, correlation of combustion parameters A~D. The vertical axis of Figure 21 is the degree of influence of each operating parameter. The combustion parameters A to D are the amounts of combustion in the burners of the respective places. If the value of the influence of the operating parameters is positive, there is a positive correlation between the observed data and the observed data. If the value of the influence of the operating parameters is negative, there is a negative correlation between the observed data and the observed data. The greater the absolute value of the value of the influence, the greater the degree of correlation between the operational parameters and the observed data. For example, if the input condition a (the amount of the raw material to be injected) is increased based on the result shown in Fig. 21, it means that the value of the observation data A also increases. Further, if the combustion parameter A is increased, it means that the value of the observation data a is decreased. The melting furnace control unit 62 refers to the observation data calculated by the observation data calculation unit 18, and changes the value of the operation state of the glass melting furnace to the above-mentioned observation data, and then changes the operation to the correlation with the observation data. parameter. Here, the operation parameter having a correlation with the observation data means an operation parameter in which, for example, the absolute value of the degree of influence on the observation data is equal to or greater than a predetermined value. For example, when the value of the observation data exceeds the upper limit value and is excessively high, the value of the operational parameter having a positive correlation with the observation data is reduced or the operation is negatively correlated with the observation data. The value of the parameter increases. Further, for example, when the value of the observation data does not reach the lower limit value and is excessively low, the value of the operational parameter having a positive correlation with the observation data is increased or is negative between the observation data and the observation data. The value of the relevant operating parameters is reduced. As a specific example, when it is determined that there is a negative correlation between the occupancy rate of the batch pile as the observation material and the furnace temperature as the operation parameter, and the occupancy rate of the batch pile exceeds the upper limit value, the melting furnace control mechanism 62 In order to increase the temperature inside the furnace, 1641 丨9. Doc • 54· 201247577 You can operate the glass melting furnace. That is, it is sufficient to increase the heating power of the burner. The process of changing the operating parameters of the melting furnace control unit 62 is equivalent to the melting furnace control step. Further, the melting furnace control unit 62 may output an alarm when the value of the observation data exceeds the upper limit value or does not reach the lower limit value. Further, the change of the operating parameters of the glass melting furnace may be performed by the operator. In this case, the melting furnace control unit 62 may not be included. Further, in this case, the operator refers to the degree of influence between the observation data calculated by the observation data calculation unit 18 and the observation data calculated by the observation data analysis unit 61 and the operation parameters, and determines how to change which operation. The parameters are fine. According to the present embodiment, since the observation data analysis unit 61 calculates the degree of influence indicating the degree of correlation of the operation parameters with respect to the observation data, it is possible to adjust the operation parameter of the glass (four) furnace according to the state of the monitored batch pile. Clarify. Further, by providing the melting furnace control mechanism 62, it is possible to automatically control the glass melting furnace to an exact state without the operator. The monthly nephew indicates that the observation data analysis unit 61 calculates the degree of influence of the operation parameters on the observation data. In addition, when the quality data (for example, the number of bubbles, etc.) indicating the quality of the raw material state is obtained, the observation data analysis unit 61 can also calculate the influence of the degree of correlation between the observation material or the operation parameter with respect to the quality ^. degree. This degree of influence can also be performed by, for example, main formation = analysis and multivariate analysis. Furthermore, the more the number of bubbles, the worse the state of the furnace. 164119. Doc -55- 201247577 Figure 22 shows the calculation of observation data a, B and the temperature a as the operating parameter. A graph of the results of the influence of ~D on the number of bubbles of one quality data. The observation data A and B are data obtained by the observation data calculation unit 18 based on the binarized image generated based on the image of the image. The temperatures A to D are values obtained by measuring the temperatures of the respective places of the glass melting furnace. In the example shown in Figure 22, if the value of the influence is positive, there is a positive correlation between the observed data or the temperature and the quality data. If the value of the influence is negative, the data or temperature and quality are also observed. There is a negative correlation between the data. Also, the larger the absolute value of the value of the influence degree, the greater the degree of correlation. For example, according to the results shown in Fig. 22, it is understood that the larger the value of the observation data a and B is, the more the number of bubbles increases (the quality deteriorates) β, and the lower the value of the temperature A, the more the number of bubbles increases. Furthermore, even under certain conditions, it is determined that there is a correlation between an observation data and quality data, and sometimes it is determined that there is a correlation between other observation materials and the above-mentioned quality data under other conditions. Fig. 23 is a graph showing changes in the state of loss or re-emergence of observation data and quality data. The vertical axis on the left side shown in Fig. 23 indicates the value of the observation data. The vertical axis on the right side is the value of the quality data (here is the number of bubbles). The horizontal axis represents the passage of time. In the example shown in Fig. 23, until the middle of the measurement period, correlation is confirmed between the observation data a and the quality data, and in the latter half, the above correlation is lost. Further, until the middle of the measurement period, there is no correlation between the observation data B and the quality data, and in the latter half, the observation data B is correlated with the quality data. Therefore, the observation data analysis unit 61 preferably calculates the observation data repeatedly and 164119. Doc •56· 201247577 The degree of influence between quality data. In the second embodiment, the degree of influence calculated by the observation data analysis unit 61 is determined, and the operation parameter related to the observation data is determined, and the operation parameter is changed based on the observation data. "The operator refers to the binarized image. 'When it is determined which operating parameter should be operated, the operator can also increase or decrease the operating parameters by referring to the binarized image. For example, when it is judged from the binarized image that the melting delay of the batch pile on the right side is observed from the upstream wall, the operator can also observe the fire of the right burner from the upstream wall. Further, in the above-described embodiments, the camera 1 may be disposed at a position from the upper right imaging fixing region 9a, and the camera iib may be disposed at a position from the upper right imaging fixing region 9b. In this case, the characteristic is The object (for example, a side wall, a burner, etc.) is also included in the imaging range, and a reference pattern or a reference point is also imaged. Thus, the camera 11a is disposed at a position from the upper right imaging fixing area 9a, and the camera 11b is disposed at a position When the position of the image fixing area % is directly above, the viewpoint conversion processing for changing the viewpoint to the fixed area \ or the fixed area 9b may not be performed. That is, the viewpoint of step S12 'S13 (refer to FIG. 9) may not be performed. In addition, in the process (see FIG. 19) which is a modification of the embodiment, the viewpoint conversion process of step S33 is not performed. [Embodiment 3] Next, as a third embodiment of the present invention A method for producing a glass article will be described. In the method for producing a glass article of the present invention, the glass melting furnace described in the second embodiment is applied. Depending on the method. Furthermore, also 164,119. Doc • 57· 201247577 The determination of the degree of correlation between the observation data and the operation parameters described in the second embodiment and the processing of changing the operation parameters are applied to the method for producing a glass article of the present invention. Fig. 24 is a schematic view showing an example of a manufacturing line of a glass article used in the method for producing a glass article of the embodiment. Further, in Fig. 24, the cameras 11a and 11b and the image processing device 13 are omitted. The cameras 11a and 11b are disposed in the vicinity of the glass melting furnace 1. Further, an image processing device 13 is also disposed. However, the configuration position of the image processing apparatus 13 is not limited. Further, the image processing device 13a described in the second embodiment can be arranged. In the manufacturing line of the glass article, a glass melting furnace 1 and a clarification tank 30 are provided. Further, the type of the clarification tank 30 is not limited. The clarification tank 3〇 may be a decompression type clarification tank in which the inside of the tank is decompressed to remove bubbles. Alternatively, the clarification tank 30 may be a high-temperature type clarification tank which removes bubbles from the inside of the tank at a high temperature. The glass melting furnace 1 (see Figs. 24 and 1) melts the glass raw material and changes it into molten glass 71. In Fig. 24, the illustration of the batch heap is omitted. The clarification tank 30 removes bubbles generated in the molten glass 71. The bubble-removed molten glass system is transferred to a forming step and a slow cooling step. Fig. 25 is a flow chart showing an example of a method for producing a glass article of the embodiment. First, the glass raw material is put into the glass melting furnace 1. The glass melting furnace 1 includes a burner 5 (see Fig. 1) to maintain the inside of the glass melting furnace 1 at a high temperature. Then, the molten glass 71 is produced by heating the glass raw material in the glass melting furnace 1 (step S91, glass melting step). In step S91, the cameras 11a, 11b image the inside of the glass melting furnace 1, and the image processing device 13 performs the image obtained by the result with the i 164119. Doc • 58 · 201247577 The same implementation. In other words, steps S51 to S54 (see Fig. 5), steps S1 to S6 (see Fig. 8), and steps S10 to S16 (see Fig. 9 or Fig. 19) are performed, such as steps S21 to S28 (see Fig. 10). By this treatment, observation information is obtained, and the inside of the glass melting furnace 1 can be well monitored. Further, in the image processing apparatus 13a described in the second embodiment, the degree of correlation between the observation data and the operating parameters of the glass melting furnace 1 can be determined in the same manner as in the second embodiment, and the operating parameters of the glass melting furnace 1 can be changed. . The molten glass 71 produced in the step S91 flows to the clarification tank 3, and bubbles are present in the molten glass 71, and a bubble layer (not shown) is formed on the surface of the molten glass 71. Inside the clarification tank 30, the bubbles of the molten glass 71 are removed (step S92, clarification step). After the step S92, the molten glass from which the bubbles are removed is formed (step S93, forming step). In the forming step, for example, the molten glass may be formed by a floating method. Specifically, 'the molten glass 71 from which the bubbles are removed is floated on the molten tin (not shown) and advanced in the transport direction to form a continuous strip-shaped glass ribbon. The glass ribbon is pressed, and the roller which is pressed against the both sides of the glass ribbon is stretched to the outside in the width direction (the direction at right angles to the conveyance direction). Then, the glass ribbon formed in step S93 is slowly cooled (step S94, slow cooling step). In the slow cooling step, the glass ribbon was pulled out from the molten tin, and the glass ribbon was slowly cooled inside the slow cooling furnace (not shown). That is, it is easy to carry it to the outside of the slow cooling furnace, and then the glass belt is slowly cooled to about the normal temperature. / After the slow cooling step, the glass ribbon solidified in the slow cooling step is processed as needed (step S95, processing step). As a processing example in the step milk, 164119. Doc • 59- 201247577 may, for example, be cut or ground. However, it is not limited to cutting or grinding, and other processing may be performed. According to the method for producing a glass article of the present embodiment, the observation of the fixing region in the glass melting furnace can be favorably maintained, and the glass article can be produced. In particular, in the same manner as in the second embodiment, the image processing device 13a determines the degree of correlation between the observation data and the operating parameters of the glass melting furnace 1, and changes the operating parameters of the glass melting furnace 1 to observe the results in the furnace. The operation of the glass-melting furnace 1 is carried out in accordance with the exact operating parameters, and the glass article is manufactured. The above description has been described in detail with reference to the specific embodiments. It is understood that the invention can be applied without departing from the spirit and scope of the invention. Multiple changes or corrections. This is based on the Japanese patent application filed on May 6, 2011 (Japanese Patent Application No. 2011-103601), the contents of which are hereby incorporated by reference. [Industrial Applicability] The present invention is preferably applied to a glass melting furnace monitoring system for monitoring a batch pile in a glass melting furnace. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a plan view showing an example of a glass gallery and a melting furnace to which the monitoring system of the glass cooling furnace of the present invention is applied. Fig. 2 is a block diagram showing a configuration example of the monitoring system in the glass melting furnace according to the first embodiment of the present invention. Fig. 3 is an explanatory view showing an example of a captured image obtained by the camera 1 la. 1641l9. Doc • 60· 201247577 Figs. 4(a) and 4(b) are explanatory diagrams showing an example of an image of a reference pattern and a matching example using a reference pattern. Fig. 5 is a flow chart showing an example of the gesture estimation operation performed by the posture specifying unit 14. Fig. 6 is a schematic view showing the range of the liquid level in accordance with the dissolved raw material in the captured image obtained by the camera 1 1 & Fig. 7 is an explanatory diagram showing an example of a conversion result obtained by converting a viewpoint to a position directly above the fixed region 93. Fig. 8 is a flow chart showing an example of the processing of the posture determination processing of the camera. Fig. 9 is a flow chart showing an example of the processing up to the observation of the data. Fig. 10 is a flow chart showing an example of the processing of the background image creation processing (step Sii). Figure 11 is a histogram obtained as a result of step S24. Figure 12 is a histogram obtained as a result of step S24. Fig. 13 is an explanatory diagram showing an example of an image after the conversion in step S13. Fig. 14 is an explanatory diagram showing an example of a background image after the conversion in step S12. Fig. 15 is an explanatory view showing an example of an image obtained as a result of the process of step S14. Fig. 16 is an explanatory diagram showing an example of an image after binarization processing. Fig. 17 is an explanatory view showing a region obtained by dividing the fixing regions 9a, 9b into a region on the side of the side wall 6 and a region on the center side of the glass melting furnace. Fig. 18 is a view showing the inside of a glass melting furnace in a modification of the first embodiment. Doc -61 ^ 201247577 Block diagram of the structure of the system. Figure 19 shows the first! A flowchart of an example of the process of processing until the observation of the data is performed in the modification of the embodiment. Fig. 20 is a block diagram showing a configuration example of a glass melting furnace monitoring system according to a second embodiment of the present invention. Fig. 21 is a graph showing an example of the result of calculating the influence degree of the operation parameter on one observation data. Fig. 22 is a graph showing the results of calculating the influence of observation data a, B and temperature A to D on one quality data. Fig. 23 is a graph showing changes in the state of loss or re-emergence of observation data and quality data. . Fig. 24 is a schematic view showing an example of a manufacturing line of a glass article used in the method for producing a glass article according to the third embodiment. Fig. 25 is a flow chart showing an example of a method for producing a glass article according to a third embodiment. [Explanation of main component symbols] 2 3a 3b 4 5 6 7 Glass melting furnace Observation window Injection inlet Injection port Exhaust burner Side wall Upstream wall 164119. Doc -62- 201247577 8 downstream wall 9a fixed area 9b fixed area 10 batch pile 11a camera lib camera 12 image memory mechanism 13 image processing device 13a image processing device 14 posture specific mechanism 15 background image forming mechanism 16 image Calibration mechanism 17 Difference calculation mechanism 18 Observation data calculation mechanism 19 Preprocessing mechanism 21a Point 21b Point 21〇 Point 21c Point 21e Point 25 Liquid level 30 Clarification groove 31a Range 41 Area 164119. Doc . 63· 201247577 42 Area 51 Area 52 Area 61 Observation data analysis mechanism 62 Molten furnace control unit 71 Molten glass 81 Part of SI in the captured image Step S2 Step S3 Step S4 Step S5 Step S6 Step S10 Step S11 Step S12 Step S13 Step S14 Step S15 Step S16 Step S21 Step S22 Step S24 Step S25 Step 164119. Doc 201247577 526 527 528 531 532 533 551 552 553 554 591 592 593 594 595 Step Step Step Step Step Step Step Step Step Step Step Step Step Step Step 164119. Doc -65