TWI722332B - Pedestrian detection method and related monitoring camera - Google Patents

Pedestrian detection method and related monitoring camera Download PDF

Info

Publication number
TWI722332B
TWI722332B TW107140120A TW107140120A TWI722332B TW I722332 B TWI722332 B TW I722332B TW 107140120 A TW107140120 A TW 107140120A TW 107140120 A TW107140120 A TW 107140120A TW I722332 B TWI722332 B TW I722332B
Authority
TW
Taiwan
Prior art keywords
detection
detection window
pedestrian
window
technology
Prior art date
Application number
TW107140120A
Other languages
Chinese (zh)
Other versions
TW202018583A (en
Inventor
陳聖元
游舜勛
劉誠傑
Original Assignee
晶睿通訊股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 晶睿通訊股份有限公司 filed Critical 晶睿通訊股份有限公司
Priority to TW107140120A priority Critical patent/TWI722332B/en
Publication of TW202018583A publication Critical patent/TW202018583A/en
Application granted granted Critical
Publication of TWI722332B publication Critical patent/TWI722332B/en

Links

Images

Landscapes

  • Closed-Circuit Television Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

A pedestrian detection method is applied to a monitoring camera. The pedestrian detection method includes forming a first detecting window on at least one monitoring frame via object analysis, utilizing human detection to adjust the first detecting window for forming a second detecting window, analyzing the second detecting window via local detection to mark an upper detecting window about a pedestrian on the monitoring frame, and determining whether to amend the second detecting window via analysis of the upper detecting window.

Description

行人偵測方法及其相關監控攝影機 Pedestrian detection method and related monitoring camera

本發明係提供一種行人偵測方法及其相關監控攝影機，尤指一種兼具低運算量與較佳偵測效能優點的高準確度之行人偵測方法及其相關監控攝影機。 The present invention provides a pedestrian detection method and related surveillance cameras, in particular a high-accuracy pedestrian detection method and related surveillance cameras that have the advantages of low computational load and better detection performance.

傳統的監控攝影機會單獨使用人臉偵測或是人形偵測掃描整張監控畫面影像以偵測和追蹤監控範圍內的行人，因此需要很高的運算量。且監控攝影機安裝就定位後，行人在監控影像內可以是任意位置或任意角度或任意行進方向，被拍攝到的圖案可能出現完整的全身影像、也可能是被其他行人或建築物或車輛遮蔽的部分身形影像，難以持續性地長期追蹤特定人臉或人形，故傳統監控攝影機的偵測和追蹤的精準度受環境影響而有所侷限。 Traditional surveillance cameras use face detection or human form detection alone to scan the entire surveillance screen image to detect and track pedestrians within the surveillance range, which requires a high amount of computation. And after the surveillance camera is installed and positioned, the pedestrian can be anywhere in the surveillance image, at any angle, or in any direction of travel. The captured pattern may appear as a complete full-body image, or it may be hidden by other pedestrians, buildings, or vehicles. Part of the body image is difficult to track a specific face or figure continuously for a long time. Therefore, the accuracy of the detection and tracking of traditional surveillance cameras is limited by the influence of the environment.

本發明係提供一種兼具低運算量與較佳偵測效能優點的高準確度之行人偵測方法及其相關監控攝影機，以解決上述之問題。 The present invention provides a high-accuracy pedestrian detection method with the advantages of low computational load and better detection performance and its related surveillance camera to solve the above-mentioned problems.

本發明之申請專利範圍係揭露一種行人偵測方法，其包含有於至少一張監控畫面進行物件分析以劃設一第一偵測視窗，利用人形偵測技術收斂該 第一偵測視窗而形成一第二偵測視窗，在該第二偵測視窗內執行人體局部偵測技術，以於該監控畫面標記一行人之一上端偵測視窗，以及分析該上端偵測視窗以決定是否修正該第二偵測視窗。 The scope of patent application of the present invention discloses a pedestrian detection method, which includes performing object analysis on at least one monitoring screen to define a first detection window, and using human form detection technology to converge the The first detection window forms a second detection window. The human body part detection technology is implemented in the second detection window to mark an upper detection window of a line of people on the monitoring screen, and analyze the upper detection Window to determine whether to modify the second detection window.

本發明之申請專利範圍另揭露一種具有行人偵測與追蹤功能的監控攝影機，其包含有一影像擷取器以及一運算處理器。該影像擷取器用來取得包含至少一張監控畫面的一影像串流。該運算處理器電連接於該影像擷取器，用來於該至少一張監控畫面進行物件分析以劃設一第一偵測視窗，利用人形偵測技術收斂該第一偵測視窗而形成一第二偵測視窗，在該第二偵測視窗內執行人體局部偵測技術，以於該監控畫面標記一行人之一上端偵測視窗，以及分析該上端偵測視窗以決定是否修正該第二偵測視窗。 The patent application scope of the present invention also discloses a surveillance camera with pedestrian detection and tracking functions, which includes an image capturer and an arithmetic processor. The image capture device is used to obtain an image stream including at least one monitoring screen. The arithmetic processor is electrically connected to the image capture device, and is used to perform object analysis on the at least one monitoring screen to define a first detection window, and use human form detection technology to converge the first detection window to form a The second detection window implements human body part detection technology in the second detection window to mark the upper detection window of a group of people on the monitoring screen, and analyze the upper detection window to determine whether to correct the second Detection window.

為了節約監控攝影機的運算效能，本發明的行人偵測方法先利用物件分析將監控畫面內的可能搜索範圍縮小為第一偵測視窗，然後利用人形偵測技術將其收斂為第二偵測視窗。第二偵測視窗已經約略相當於行人在監控畫面的尺寸，行人偵測方法僅需於第二偵測視窗(或其內部的局部偵測範圍)裡執行精密的人頭偵測技術或人臉偵測技術，找出行人頭部輪廓或臉部特徵，既能節約監控攝影機的運算能力、也可即時得到偵測結果而不致延誤。此外，行人的頭部輪廓或臉部特徵還能用來修正人形偵測結果，修正後的人形偵測結果再用來找出其精確的行人頭部輪廓或臉部特徵，此流程不斷地重複反饋，逐漸辨識出正確的行人位置、及分辨重疊或緊靠的人群。無論行人處於任意角度或任意位置，本發明的監控攝影機都能以低階硬體配備快速取得高精準度的行人偵測結果。 In order to save the computing performance of the surveillance camera, the pedestrian detection method of the present invention first uses object analysis to narrow the possible search range in the surveillance screen to the first detection window, and then uses the human form detection technology to converge it to the second detection window . The second detection window is roughly equivalent to the size of the pedestrian in the monitoring screen. The pedestrian detection method only needs to implement sophisticated head detection technology or face in the second detection window (or its internal partial detection range) The detection technology can find out the head contour or facial features of pedestrians, which not only saves the computing power of the surveillance camera, but also obtains the detection results in real time without delay. In addition, the pedestrian’s head profile or facial features can also be used to correct the human form detection results, and the corrected human form detection results can then be used to find out the exact pedestrian’s head profile or facial features. This process is repeated continuously Feedback, and gradually identify the correct pedestrian position, and distinguish overlapping or close crowds. No matter where the pedestrian is at any angle or any position, the surveillance camera of the present invention can quickly obtain high-precision pedestrian detection results with low-level hardware.

10:監控攝影機 10: Surveillance camera

12:影像擷取器 12: Image grabber

14:運算處理器 14: arithmetic processor

16:資料庫 16: Database

W1:第一偵測視窗 W1: The first detection window

W2:第二偵測視窗 W2: Second detection window

W3:第三偵測視窗 W3: Third detection window

Wh、Wh’:上端偵測視窗 Wh, Wh’: Upper detection window

R:局部偵測範圍 R: Local detection range

S200、S202、S204、S206、S208、S210、S212、S214、S216、S218、S220:步驟 S200, S202, S204, S206, S208, S210, S212, S214, S216, S218, S220: steps

第1圖為本發明實施例之監控攝影機之功能方塊圖。 Figure 1 is a functional block diagram of a surveillance camera according to an embodiment of the present invention.

第2圖為本發明實施例之行人偵測方法之流程圖。 Figure 2 is a flowchart of a pedestrian detection method according to an embodiment of the present invention.

第3圖至第7圖分別為本發明實施例之監控畫面於不同偵測階段的變化示意圖。 Figures 3 to 7 are respectively schematic diagrams of changes of the monitoring screen in different detection stages according to the embodiment of the present invention.

第8圖為本發明另一實施例之行人偵測方法之流程圖。 FIG. 8 is a flowchart of a pedestrian detection method according to another embodiment of the present invention.

請參閱第1圖，第1圖為本發明實施例之監控攝影機10之功能方塊圖。第2圖所述之行人偵測方法適用於第1圖所示之監控攝影機10。監控攝影機10可包含影像擷取器12以及運算處理器14。監控攝影機10裝設在待監控區域內。影像擷取器12用來取得關聯於監控區域的影像串流，且影像串流包含多張監控畫面。運算處理器14電連接於影像擷取器12。運算處理器14利用影像串流中的至少一張監控畫面執行本發明的行人偵測方法；或者，也可比較複數張監控畫面間的參數差異以執行行人偵測方法，且複數張監控畫面可為連續或非連續的監控畫面，例如影像串流中共有十張監控畫面，連續的監控畫面係指由第一張依順序取到第十張監控畫面以執行行人偵測方法，此外，於另一實施例中，不連續的監控畫面係指當該十張監控畫面中有部分監控畫面的影像品質不佳(例如於監控畫面中行人於監控畫面中的比例小於一預設條件、人臉因為拍攝角度而無法分析、或是人臉模糊等)，則該些監控畫面可捨棄，而只取剩餘的監控畫面來執行行人偵測方法。 Please refer to FIG. 1, which is a functional block diagram of the surveillance camera 10 according to an embodiment of the present invention. The pedestrian detection method described in FIG. 2 is applicable to the surveillance camera 10 shown in FIG. 1. The surveillance camera 10 may include an image capturer 12 and an arithmetic processor 14. The monitoring camera 10 is installed in the area to be monitored. The image capture device 12 is used to obtain an image stream associated with the monitoring area, and the image stream includes multiple monitoring pictures. The computing processor 14 is electrically connected to the image capturer 12. The arithmetic processor 14 uses at least one monitoring screen in the image stream to execute the pedestrian detection method of the present invention; or, it can also compare the parameter difference between a plurality of monitoring screens to perform the pedestrian detection method, and the plurality of monitoring screens can be It is a continuous or non-continuous monitoring screen. For example, there are ten monitoring screens in the video stream. The continuous monitoring screen refers to taking the first to the tenth monitoring screen in order to perform the pedestrian detection method. In one embodiment, the discontinuous monitoring picture refers to when the image quality of some of the ten monitoring pictures is poor (for example, the ratio of pedestrians in the monitoring picture in the monitoring picture is less than a preset condition, and the face is caused by If the shooting angle cannot be analyzed, or the face is blurred, etc.), the monitoring images can be discarded, and only the remaining monitoring images are used to implement the pedestrian detection method.

請參閱第2圖與第3圖至第7圖，第2圖為本發明實施例之行人偵測方法之流程圖，第3圖至第7圖分別為本發明實施例之監控畫面I於不同偵測階段的 變化示意圖。首先，執行步驟S200，行人偵測方法在監控畫面I進行物件分析以劃設第一偵測視窗W1。物件分析可包含移動偵測技術、或物件追蹤技術、或前述兩者技術之組合、或任意可於影像畫面中分析出物件所在之方法皆屬於本發明之應用範疇。舉例來說，影像串流具有多張監控畫面，行人偵測方法可於第一張監控畫面使用移動偵測劃設第一偵測視窗W1，並於第二張監控畫面使用物件追蹤劃設第一偵測視窗W1，意即不同監控畫面可任意選用不同物件分析技術；或者，行人偵測方法可於所有監控畫面都使用移動偵測劃設第一偵測視窗W1，意即所有監控畫面皆使用同一類型的物件分析技術。目的在於可依據特定因素(例如影像品質或其它參數)決定監控畫面套用何種類型的物件分析技術，以取得最佳行人偵測結果。 Please refer to Fig. 2 and Fig. 3 to Fig. 7. Fig. 2 is a flowchart of a pedestrian detection method according to an embodiment of the present invention, and Figs. 3 to 7 are respectively different monitoring screens of an embodiment of the present invention. Detection phase Schematic diagram of changes. First, in step S200, the pedestrian detection method performs object analysis on the monitoring screen I to define a first detection window W1. Object analysis can include motion detection technology, or object tracking technology, or a combination of the foregoing two technologies, or any method that can analyze the location of the object in the image frame, which belongs to the application scope of the present invention. For example, if the image stream has multiple monitoring screens, the pedestrian detection method can use motion detection to set the first detection window W1 on the first monitoring screen, and use the object tracking to set the first detection window on the second monitoring screen. One detection window W1, which means that different object analysis technologies can be selected for different monitoring screens; or, the pedestrian detection method can use motion detection on all monitoring screens. Set up the first detection window W1, which means that all monitoring screens are Use the same type of object analysis technology. The purpose is to determine which type of object analysis technology to apply to the monitoring screen based on specific factors (such as image quality or other parameters) in order to obtain the best pedestrian detection results.

移動偵測係可將監控畫面I區分為多個小範圍區域，通過比較前後兩張監控畫面I在相同位置的小範圍區域是否有亮度等參數變化，判斷那些小範圍區域內有物件移動的情形發生，該些小範圍區域即定義為第一偵測視窗W1，如第3圖所示之多個小矩形及其組合成的不規則形狀。移動偵測的所得結果較為概略，並非標記出行人的精準輪廓。此外，移動偵測技術的實施方式不限於前揭態樣所述，端視設計需求而定。物件追蹤係以視覺為基礎的物件追蹤法則，利用過去監控畫面的物件所在位置當作起始點，利用慣性移動特性預測目標物件位置，或計算目前監控畫面的周圍差異值的大小，逐漸往遞減的方向移動直到差異值小於門檻值為止，以預測目標物件位置。物件追蹤的所得結果並非行人的精準位置和大小輪廓。此外，物件追蹤技術的實施方式不限於前揭態樣所述，端視設計需求而定。 The motion detection system can divide the monitoring screen I into multiple small areas. By comparing the two monitoring screens I before and after the small area at the same position, whether there are parameter changes such as brightness in the small area, it can judge the situation of objects moving in those small areas. When it happens, these small areas are defined as the first detection window W1, as shown in FIG. 3, a plurality of small rectangles and their combined irregular shapes. The result of motion detection is more general, not marking the precise contour of pedestrians. In addition, the implementation of the motion detection technology is not limited to that described in the previous disclosure, and it depends on the design requirements. Object tracking is a vision-based object tracking law that uses the location of the object in the past monitoring screen as the starting point, uses inertial movement characteristics to predict the location of the target object, or calculates the size of the difference value around the current monitoring screen, gradually decreasing Move in the direction of until the difference value is less than the threshold value to predict the location of the target object. The result of object tracking is not the precise location and size contour of pedestrians. In addition, the implementation of the object tracking technology is not limited to those described in the previous disclosure, and depends on the design requirements.

接著，執行步驟S202，利用人形偵測技術收斂第一偵測視窗W1而形 成第二偵測視窗W2，如第4圖所示。一般來說，人形偵測技術係根據背景亮度變化或視差角度等因素，提取產生亮度梯度的邊緣，得到可能的行人範圍，意即第二偵測視窗W2。人形偵測技術亦不限於前揭實施態樣所述，端視設計需求而定。然後，執行步驟S204與步驟S206，在第二偵測視窗W2內估算且劃設局部偵測範圍R，以於局部偵測範圍R內執行人體局部偵測技術，標記行人的上端偵測視窗Wh，如第5圖及第6圖所示。局部偵測範圍R之劃設係為了進一步限縮第二偵測視窗W2的尺寸。人體局部偵測技術不需分析範圍較大的第二偵測視窗W2，僅需分析範圍較小的局部偵測範圍R，而能有效提高執行效率。因此，步驟S204係為選擇性手段，行人偵測方法也可以不劃設局部偵測範圍R，直接在第二偵測視窗W2內執行人體局部偵測技術。 Then, step S202 is executed to converge the first detection window W1 by using the human form detection technology to form a shape A second detection window W2 is formed, as shown in Fig. 4. Generally speaking, the human form detection technology extracts the edge that generates the brightness gradient based on the background brightness change or the parallax angle and other factors to obtain the possible pedestrian range, which is the second detection window W2. The humanoid detection technology is not limited to the implementation mode described in the previous disclosure, and it depends on the design requirements. Then, step S204 and step S206 are performed to estimate and set a partial detection range R in the second detection window W2, so as to implement the human body partial detection technology in the partial detection range R, and mark the upper detection window Wh of the pedestrian , As shown in Figure 5 and Figure 6. The partial detection range R is designed to further limit the size of the second detection window W2. The human body part detection technology does not need the second detection window W2 with a larger analysis range, and only needs the local detection range R with a smaller analysis range, which can effectively improve the execution efficiency. Therefore, step S204 is an optional method, and the pedestrian detection method can also directly execute the human body partial detection technology in the second detection window W2 without setting the local detection range R.

特別一提的是，本發明的人體局部偵測技術包含(但不限於)人頭偵測技術與人臉偵測技術。人頭偵測技術係可取得第二偵測視窗W2內的色階變化度累積值，據此定義行人之頭部尺寸與位置。人臉偵測技術係可抽取第二偵測視窗W2內的臉部特徵值，據此定義行人之臉部尺寸與位置。由於人頭偵測技術和人臉偵測技術所需的運算量較大，故本發明先利用物件分析以及人形偵測技術縮小行人在監控畫面I內可能出現的區域，再由人體局部偵測技術針對限縮後範圍進行更為精密的偵測與分析。第3圖所示之人形偵測技術係偵測行人的全身範圍；基於上述減少運算量之理由，人形偵測技術也可以只偵測行人的上半身範圍(未繪製於圖式中)，再由人體局部偵測技術找出行人的頭部輪廓或臉部特徵。 In particular, the human body part detection technology of the present invention includes (but is not limited to) human head detection technology and human face detection technology. The human head detection technology can obtain the cumulative value of the gradation change in the second detection window W2, and define the size and position of the pedestrian's head accordingly. The face detection technology can extract the facial feature values in the second detection window W2, and define the size and position of the pedestrian's face accordingly. Since the human head detection technology and the human face detection technology require a large amount of computation, the present invention first uses object analysis and human figure detection technology to reduce the area where pedestrians may appear in the monitoring screen I, and then detects the human body part. The technology performs more precise detection and analysis for the narrowed range. The humanoid detection technology shown in Figure 3 detects the full-body range of pedestrians; based on the aforementioned reasons for reducing the amount of calculation, the humanoid detection technology can also detect only the upper body range of pedestrians (not shown in the diagram), and then Human body part detection technology finds out the head contour or facial features of pedestrians.

再來，執行步驟S208，行人偵測方法分析上端偵測視窗Wh，判斷是否修正第二偵測視窗W2。例如上端偵測視窗Wh的所在區位於第二偵測視窗W2的預估頭部範圍，表示第二偵測視窗W2確實符合被測行人的位置與大小，執行 步驟S210，不需進行反饋修正。如果上端偵測視窗Wh和第二偵測視窗W2的預估頭部範圍差距甚遠，執行步驟S212，根據上端偵測視窗Wh與第二偵測視窗W2間的座標值差異修正，第二偵測視窗W2修正後即生成第三偵測視窗W3，如第7圖所示。將第4圖所示第二偵測視窗W2與第7圖所示第三偵測視窗W3相比，可看出第二偵測視窗W2係重疊於兩個行人的圖像，修正後的第三偵測視窗W3則往右偏，而能精確標記右側行人。 Then, step S208 is executed, and the pedestrian detection method analyzes the upper detection window Wh to determine whether to modify the second detection window W2. For example, the area where the upper detection window Wh is located is within the estimated head range of the second detection window W2, which means that the second detection window W2 is indeed in line with the position and size of the pedestrian under test. In step S210, no feedback correction is required. If the estimated head range of the upper detection window Wh and the second detection window W2 is far apart, step S212 is executed to correct the difference in coordinate values between the upper detection window Wh and the second detection window W2, and the second detection After the window W2 is corrected, a third detection window W3 is generated, as shown in FIG. 7. Comparing the second detection window W2 shown in Figure 4 with the third detection window W3 shown in Figure 7, it can be seen that the second detection window W2 overlaps the images of the two pedestrians, and the corrected first The three detection window W3 is shifted to the right, and can accurately mark pedestrians on the right.

本發明的行人偵測方法還可在第三偵測視窗W3內另行執行人體局部偵測技術，並相應標記上端偵測視窗Wh’，意即執行步驟S214。行人偵測方法利用上端偵測視窗Wh’分析第三偵測視窗W3是否確實符合被測行人，若不符則依照前揭步驟S208~S212，判斷如何修正第三偵測視窗W3。由此可知，行人偵測方法先以低運算量的物件分析與人形偵測技術估算行人在監控畫面I裡面的大致範圍，並選擇性劃設局部偵測範圍以限縮行人在監控畫面I裡面的可能範圍，然後才運用精準度較佳但高運算量的人體局部偵測技術標記行人臉部或頭部。除此之外，行人的臉部或頭部被標記後，還能用來分析判斷前述的可能範圍是否精確，如不精確便可再次進行修正。換句話說，行人的整體輪廓(全身或半身)與局部輪廓(臉部或頭部)會重覆地來回比對反饋，逐漸收斂出最佳的偵測結果。 The pedestrian detection method of the present invention can additionally implement the human body partial detection technology in the third detection window W3, and mark the upper detection window Wh' accordingly, which means that step S214 is performed. The pedestrian detection method uses the upper detection window Wh' to analyze whether the third detection window W3 actually matches the detected pedestrian, and if it does not, it determines how to correct the third detection window W3 according to the steps S208 to S212 described above. It can be seen that the pedestrian detection method first estimates the approximate range of pedestrians in the monitoring screen I with low computational object analysis and human form detection technology, and selectively sets a partial detection range to limit the pedestrians in the monitoring screen I. Then, the human body part detection technology with better accuracy and high computational complexity is used to mark the face or head of pedestrians. In addition, after the pedestrian's face or head is marked, it can also be used to analyze and determine whether the aforementioned possible range is accurate, and if it is not accurate, it can be corrected again. In other words, the pedestrian's overall profile (full body or half-length) and partial profile (face or head) will be repeatedly compared back and forth for feedback, gradually converging to the best detection result.

步驟S208係可通過多種方式分析上端偵測視窗Wh，決定是否修正第二偵測視窗W2。舉例來說，監控攝影機10可以內建資料庫(未繪製於圖式中)、或是以有線或無線方式連線到外部資料庫16。資料庫16儲存預設的查找表，查找表具有人類身高與頭部的比例資料；例如頭部小表示身高矮，頭部大表示身材高大。故行人偵測方法可分析上端偵測視窗Wh的尺寸，根據查找表估算行人的體型尺寸，判斷第二偵測視窗W2是否符合估算的行人體型，決定是否需修正 第二偵測視窗W2。再者，行人偵測方法還可選擇以上端偵測視窗Wh再次執行該人形偵測技術，意即以上端偵測視窗Wh作為行人的臉部或頭部位置，生成人形偵測結果(輪廓或是範圍)，然後比較人形偵測結果是否相符第二偵測視窗W2，決定是否要修正第二偵測視窗W2。又或者，行人偵測方法另能將上端偵測視窗Wh定義為行人的臉部或頭部中心，依上端偵測視窗Wh的座標值以預定比例直接移動第二偵測視窗W2，使第二偵測視窗W2相對於上端偵測視窗Wh達到置中位置。 In step S208, the upper detection window Wh can be analyzed in a variety of ways to determine whether to modify the second detection window W2. For example, the surveillance camera 10 may have a built-in database (not shown in the drawing), or be connected to the external database 16 in a wired or wireless manner. The database 16 stores a preset look-up table, and the look-up table has ratio data of human height to head; for example, a small head means short height, and a large head means tall body. Therefore, the pedestrian detection method can analyze the size of the upper detection window Wh, estimate the size of the pedestrian according to the look-up table, determine whether the second detection window W2 meets the estimated pedestrian shape, and determine whether it needs to be corrected The second detection window W2. Furthermore, the pedestrian detection method can also select the upper detection window Wh to execute the human figure detection technique again, which means that the upper detection window Wh is used as the position of the pedestrian’s face or head to generate the human figure detection result (contour or Yes range), and then compare whether the human figure detection result matches the second detection window W2, and decide whether to modify the second detection window W2. Or, the pedestrian detection method can additionally define the upper detection window Wh as the center of the pedestrian's face or head, and directly move the second detection window W2 in a predetermined ratio according to the coordinate value of the upper detection window Wh, so that the second detection window W2 The detection window W2 reaches the center position relative to the upper detection window Wh.

請參閱第8圖，第8圖為本發明另一實施例之行人偵測方法之流程圖。第8圖所述之行人偵測方法適用於第1圖所示之監控攝影機10。步驟S212執行完畢後，行人偵測方法可選擇性執行步驟S216，比較第二偵測視窗W2與第三偵測視窗W3的尺寸差異。若是第二偵測視窗W2小於或等於第三偵測視窗W3、或是第二偵測視窗W2僅略大於第三偵測視窗W3，表示第三偵測視窗W3屬於校正人形偵測技術的偏差，判斷被測行人只有一位，如步驟S218。若第二偵測視窗W2大於第三偵測視窗W3的差異超出預設門檻值，判斷被測行人可能是位置重疊或彼此緊靠多個行人，如步驟S220，故行人偵測方法在第二偵測視窗W2未重疊於第三偵測視窗W3的範圍內執行人體局部偵測技術，找出監控畫面I裡面的另一個行人。接著，可接續執行步驟S204，逐步確認與收斂另一個行人的偵測視窗。 Please refer to FIG. 8, which is a flowchart of a pedestrian detection method according to another embodiment of the present invention. The pedestrian detection method described in FIG. 8 is applicable to the surveillance camera 10 shown in FIG. 1. After step S212 is executed, the pedestrian detection method can selectively execute step S216 to compare the size difference between the second detection window W2 and the third detection window W3. If the second detection window W2 is less than or equal to the third detection window W3, or the second detection window W2 is only slightly larger than the third detection window W3, it means that the third detection window W3 is a deviation of the humanoid detection technology. , It is judged that there is only one pedestrian under test, as in step S218. If the difference between the second detection window W2 and the third detection window W3 exceeds the preset threshold value, it is determined that the detected pedestrian may be overlapped or close to multiple pedestrians, such as step S220, so the pedestrian detection method is in the second The detection window W2 is not overlapped with the third detection window W3, and the human body partial detection technology is executed to find another pedestrian in the monitoring screen I. Then, step S204 can be continued to gradually confirm and converge another pedestrian detection window.

關於第2圖及第8圖所述之行人偵測方法，執行步驟S212之後，本發明的行人偵測方法可以僅執行步驟S214、也可以僅執行步驟S216、S218及S220，當然也能一併或輪流地執行步驟S214~S220，端視實際需求，由使用者自行於監控攝影機10進行設定。 Regarding the pedestrian detection method described in Figures 2 and 8, after step S212 is performed, the pedestrian detection method of the present invention may only perform step S214, or only perform steps S216, S218, and S220, of course, can also be combined Or, steps S214 to S220 are executed in turn, and the user can set it on the surveillance camera 10 according to actual needs.

為了節約監控攝影機的運算效能，本發明的行人偵測方法先利用物件分析將監控畫面內的可能搜索範圍縮小為第一偵測視窗，然後利用人形偵測技術將其收斂為第二偵測視窗。第二偵測視窗已經約略相當於行人在監控畫面的尺寸，行人偵測方法僅需於第二偵測視窗(或其內部的局部偵測範圍)裡執行精密的人頭偵測技術或人臉偵測技術，找出行人頭部輪廓或臉部特徵，既能節約監控攝影機的運算能力、也可即時得到偵測結果而不致延誤。此外，行人的頭部輪廓或臉部特徵還能用來修正人形偵測結果，修正後的人形偵測結果再用來找出其精確的行人頭部輪廓或臉部特徵，此流程不斷地重複反饋，逐漸辨識出正確的行人位置、及分辨重疊或緊靠的人群。相較先前技術，無論行人處於任意角度或任意位置，本發明的監控攝影機都能以低階硬體配備快速取得高精準度的行人偵測結果。 In order to save the computing performance of the surveillance camera, the pedestrian detection method of the present invention first uses object analysis to narrow the possible search range in the surveillance screen to the first detection window, and then uses the human form detection technology to converge it to the second detection window . The second detection window is roughly equivalent to the size of the pedestrian in the monitoring screen. The pedestrian detection method only needs to implement sophisticated head detection technology or face in the second detection window (or its internal partial detection range) The detection technology can find out the head contour or facial features of pedestrians, which not only saves the computing power of the surveillance camera, but also obtains the detection results in real time without delay. In addition, the pedestrian’s head profile or facial features can also be used to correct the human form detection results, and the corrected human form detection results can then be used to find out the exact pedestrian’s head profile or facial features. This process is repeated continuously Feedback, and gradually identify the correct pedestrian position, and distinguish overlapping or close crowds. Compared with the prior art, no matter where the pedestrian is at any angle or any position, the surveillance camera of the present invention can quickly obtain high-precision pedestrian detection results with low-level hardware.

以上所述僅為本發明之較佳實施例，凡依本發明申請專利範圍所做之均等變化與修飾，皆應屬本發明之涵蓋範圍。 The foregoing descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made in accordance with the scope of the patent application of the present invention should fall within the scope of the present invention.

S200、S202、S204、S206、S208、S210、S212、S214:步驟 S200, S202, S204, S206, S208, S210, S212, S214: steps

Claims (12)

一種行人偵測方法，其包含有：於至少一張監控畫面內進行物件分析以劃設一第一偵測視窗；利用人形偵測技術收斂該第一偵測視窗而形成一第二偵測視窗；修正該第二偵測視窗以生成一第三偵測視窗；比較該第二偵測視窗與該第三偵測視窗之尺寸差異；該第二偵測視窗大於該第三偵測視窗時，在該第二偵測視窗未重疊該第三偵測視窗之範圍內執行該人體局部偵測技術，以於該監控畫面標記一行人之一上端偵測視窗；以及分析該上端偵測視窗以決定是否修正該第二偵測視窗。 A pedestrian detection method, comprising: performing object analysis in at least one monitoring screen to define a first detection window; using human form detection technology to converge the first detection window to form a second detection window ; Modify the second detection window to generate a third detection window; compare the size difference between the second detection window and the third detection window; when the second detection window is larger than the third detection window, Perform the human body part detection technology within the range where the second detection window does not overlap the third detection window to mark the upper detection window of a group of people on the monitoring screen; and analyze the upper detection window to determine Whether to modify the second detection window. 如請求項1所述之行人偵測方法，其中在該第二偵測視窗內執行人體局部偵測技術係包含：在該第二偵測視窗內估算且劃設一局部偵測範圍；以及於該局部偵測範圍內執行該人體局部偵測技術。 The pedestrian detection method according to claim 1, wherein executing the human body partial detection technology in the second detection window includes: estimating and setting a partial detection range in the second detection window; and The human body local detection technology is executed within the local detection range. 如請求項1所述之行人偵測方法，其中該人體局部偵測技術係為人頭偵測技術或人臉偵測技術。 The pedestrian detection method according to claim 1, wherein the human body partial detection technology is a human head detection technology or a human face detection technology. 如請求項3所述之行人偵測方法，其中該人頭偵測技術取得該第二偵測視窗內的一色階變化度累積值，據此定義該行人之一頭部尺寸與位置。 The pedestrian detection method according to claim 3, wherein the human head detection technology obtains a cumulative value of gradation change in the second detection window, and thereby defines the size and position of a head of the pedestrian. 如請求項3所述之行人偵測方法，其中該人臉偵測技術抽取該第二偵測視窗內的臉部特徵值，據此定義該行人之一臉部尺寸與位置。 The pedestrian detection method according to claim 3, wherein the face detection technology extracts facial feature values in the second detection window, and accordingly defines the size and position of a face of the pedestrian. 如請求項1所述之行人偵測方法，其中該人形偵測技術係偵測該行人之一全身範圍、或該行人之一上半身範圍。 The pedestrian detection method according to claim 1, wherein the human form detection technology detects a whole body range of the pedestrian or an upper body range of the pedestrian. 如請求項1所述之行人偵測方法，其中分析該上端偵測視窗以決定是否修正該第二偵測視窗係包含：連線至一資料庫以取得一查找表；以及根據該查找表估算該行人之體型尺寸，並相應修正該第二偵測視窗。 The pedestrian detection method according to claim 1, wherein analyzing the upper detection window to determine whether to modify the second detection window includes: connecting to a database to obtain a lookup table; and estimating based on the lookup table The body size of the pedestrian, and correct the second detection window accordingly. 如請求項1所述之行人偵測方法，其中分析該上端偵測視窗以決定是否修正該第二偵測視窗係包含：以該上端偵測視窗為基準再次執行該人形偵測技術；以及根據人形偵測結果修正該第二偵測視窗。 The pedestrian detection method according to claim 1, wherein analyzing the upper detection window to determine whether to modify the second detection window includes: re-executing the humanoid detection technology based on the upper detection window; and according to The human figure detection result modifies the second detection window. 如請求項1所述之行人偵測方法，其中分析該上端偵測視窗以決定是否修正該第二偵測視窗係包含：依照該上端偵測視窗之一座標值以一預定比例修正該第二偵測視窗之位置。 The pedestrian detection method according to claim 1, wherein analyzing the upper detection window to determine whether to correct the second detection window includes: correcting the second detection window by a predetermined ratio according to a coordinate value of the upper detection window Detect the position of the window. 如請求項1所述之行人偵測方法，另包含有：修正該第二偵測視窗以生成一第三偵測視窗；以及於該第三偵測視窗內執行該人體局部偵測技術，判斷是否需修正該第三偵測視窗。 The pedestrian detection method according to claim 1, further comprising: modifying the second detection window to generate a third detection window; and executing the human partial detection technology in the third detection window to determine Whether to modify the third detection window. 如請求項1所述之行人偵測方法，其中該物件分析包含移動偵測、 或物件追蹤、或前述兩者之組合。 The pedestrian detection method according to claim 1, wherein the object analysis includes motion detection, Or object tracking, or a combination of the two. 一種具有行人偵測功能的監控攝影機，其包含有：一影像擷取器，用來取得包含至少一張監控畫面的一影像串流；以及一運算處理器，電連接於該影像擷取器，用來執行如請求項1至11其中任一所述的行人偵測方法。 A surveillance camera with a pedestrian detection function, comprising: an image capture device for obtaining an image stream including at least one monitoring screen; and an arithmetic processor, which is electrically connected to the image capture device, It is used to implement the pedestrian detection method described in any one of claims 1 to 11.
TW107140120A 2018-11-13 2018-11-13 Pedestrian detection method and related monitoring camera TWI722332B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW107140120A TWI722332B (en) 2018-11-13 2018-11-13 Pedestrian detection method and related monitoring camera

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW107140120A TWI722332B (en) 2018-11-13 2018-11-13 Pedestrian detection method and related monitoring camera

Publications (2)

Publication Number Publication Date
TW202018583A TW202018583A (en) 2020-05-16
TWI722332B true TWI722332B (en) 2021-03-21

Family

ID=71895793

Family Applications (1)

Application Number Title Priority Date Filing Date
TW107140120A TWI722332B (en) 2018-11-13 2018-11-13 Pedestrian detection method and related monitoring camera

Country Status (1)

Country Link
TW (1) TWI722332B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201228332A (en) * 2010-12-20 2012-07-01 Hui-Chuan Chien Mobile electronic device
CN102957860A (en) * 2011-08-11 2013-03-06 三星电子株式会社 Apparatus and method for processing image
US20130076913A1 (en) * 2011-09-28 2013-03-28 Xerox Corporation System and method for object identification and tracking
US20150178293A1 (en) * 2011-02-14 2015-06-25 Microsoft Technology Licensing, Llc Change invariant scene recognition by an agent
CN106484115A (en) * 2011-10-28 2017-03-08 奇跃公司 For strengthening the system and method with virtual reality

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201228332A (en) * 2010-12-20 2012-07-01 Hui-Chuan Chien Mobile electronic device
US20150178293A1 (en) * 2011-02-14 2015-06-25 Microsoft Technology Licensing, Llc Change invariant scene recognition by an agent
CN102957860A (en) * 2011-08-11 2013-03-06 三星电子株式会社 Apparatus and method for processing image
US20130076913A1 (en) * 2011-09-28 2013-03-28 Xerox Corporation System and method for object identification and tracking
CN106484115A (en) * 2011-10-28 2017-03-08 奇跃公司 For strengthening the system and method with virtual reality

Also Published As

Publication number Publication date
TW202018583A (en) 2020-05-16

Similar Documents

Publication Publication Date Title
US10417503B2 (en) Image processing apparatus and image processing method
US9576367B2 (en) Object detection method and device
US9311719B2 (en) Image monitoring apparatus for estimating gradient of singleton, and method therefor
US9418278B2 (en) Image analysis method, camera apparatus, control apparatus, control method and storage medium
TWI640931B (en) Image object tracking method and apparatus
TWI405143B (en) Object image correcting apparatus and method of identification
JP2013537661A (en) Automatic detection of moving objects using stereo vision technology
CN103996019A (en) System and method used for detecting and tracking a plurality of portions on an object
US10643338B2 (en) Object detection device and object detection method
KR20140148171A (en) Lane Detection method for Advanced Vehicle
JP2006090957A (en) Surrounding object detecting device for moving body, and surrounding object detection method for moving body
TWI722332B (en) Pedestrian detection method and related monitoring camera
US10719707B2 (en) Pedestrian detection method and related monitoring camera
CN110322508B (en) Auxiliary positioning method based on computer vision
JP5217917B2 (en) Object detection and tracking device, object detection and tracking method, and object detection and tracking program
CN113642546B (en) Multi-face tracking method and system
CN108573497B (en) Passenger flow statistical device and method
KR20200079013A (en) Method for tracking air-targets using correlation filter and system thereof
CN111182259A (en) Pedestrian detection method and related monitoring camera thereof
KR100994366B1 (en) Method for tracking a movement of a moving target of image tracking apparatus
JP6559987B2 (en) Person authentication device and person authentication method
TWI676964B (en) Image stitching method and related monitoring camera device
JP2004240909A (en) Image processor and image processing method
JP4937318B2 (en) Image processing apparatus and image adjustment method
KR20160079511A (en) Device and method for recognizing parking stall