TWI526988B

TWI526988B - Motion variation based adaptive local area iterative updating method for background modeling

Info

Publication number: TWI526988B
Application number: TW103120132A
Authority: TW
Inventors: 吳炫陞; 林道通; 簡剛民
Original assignee: 尚茂智能科技股份有限公司; 國立臺北大學
Priority date: 2014-06-10
Filing date: 2014-06-10
Publication date: 2016-03-21
Also published as: TW201546756A

Description

基於移動變化量之適應性區域局部迭代更新之背景建立方法 Background establishment method for local iterative update of adaptive region based on mobile variation

本發明揭露一種局部背景更新的方法，特別是一種應用於影像監視系統的背景更新模組方法。 The invention discloses a method for partial background update, in particular to a background update module method applied to an image monitoring system.

在日常生活中，從複雜背景中進行移動影像物件的偵測，是一個在影像監視的系統上，非常重要且非常困難的問題。然而，目前出現很多套建立背景模組的方法，卻沒有辦法有效地分辨出是否偵測到的物件是真正的移動影像物件，且諸多的方法都是以全區域做背景更新。 In daily life, the detection of moving image objects from complex backgrounds is a very important and very difficult problem in image surveillance systems. However, there are many ways to create a background module, but there is no way to effectively distinguish whether the detected object is a real moving image object, and many methods are to update the background with the whole area.

例如美國專利8,009,918中所揭露的背景更新模組的方法，是使用隨機機率來決定每個像素的更新速度，當複雜的背景中，有許多非真正的移動影像物件，如：樹葉晃動，都是以隨機的機率來把非真正的移動物件做更新消除，無法在短時間內快速地消除；或是當物件本身就處於停留很久的狀態，之後離開原先位置卻無法很快速的消除掉留下來的偽前景資訊。 For example, the method of updating the module in the background disclosed in U.S. Patent No. 8,009,918 is to use a random probability to determine the update speed of each pixel. In a complicated background, there are many non-real moving image objects, such as: leaf shaking, Randomly eliminates non-real moving objects and can't be quickly eliminated in a short time; or when the object itself stays in a long time, then leaving the original position can't quickly eliminate the remaining ones. Pseudo-foreground information.

由此可知，在先前背景模組建立的技術中，無法很快速的解決在複雜環境下，分辨出是否偵測到的物件是真正的移動影像物件，最主要的原因就在於其中的更新模組方法。 It can be seen that in the technology established by the previous background module, it is not possible to quickly solve the problem in the complex environment to distinguish whether the detected object is a real moving image object, and the main reason is the update module. method.

為了解決上述問題，本發明提供出一種更新模組的新方法，適用於任何背景模組，能夠局部區域以不同的更新速度做背景的更新替換，改善不必要更新的背景區域以及需要加速更新的背景區域。此方法必須提供一連續影像集合，影像集合至少包含有時間序列關係的第一影像與第二影像，且第一影像早於第二影像，從連續影像集合中，依序的比較第二影像與第一影像中物體的位置差異，觀察各個區域物件出現移動情形，並且將各個區域物件出現移動情形在短暫的時間內以迭代的方式做統計，進而得知該區域物件出現移動程度，即可調整該區域的背景更新速度。 In order to solve the above problems, the present invention provides a new method for updating a module, which is applicable to any background module, can replace and replace the background with different update speeds in a local area, improve an unnecessary update background area, and needs to speed up the update. Background area. The method must provide a continuous image set, the image set includes at least a first image and a second image in a time series relationship, and the first image is earlier than the second image, and the second image is sequentially compared from the continuous image set. The difference in the position of the object in the first image, observe the movement of the objects in each area, and perform the iterative manner in a short period of time by moving the movement of each object, and then know the movement degree of the object in the area, and then adjust The background update speed of the area.

因此，本發明之主要目的在於提供出區域物件出現移動程度大小的數值來改善任何背景模組的全區域更新方式，能夠以區域物件出現移動程度大小的數值來決定任何區域做局部背景更新速度。 Therefore, the main object of the present invention is to provide a numerical value of the degree of movement of a region object to improve the overall region update mode of any background module, and to determine the local background update speed of any region by the value of the degree of movement of the region object.

利用第一影像t-1與第二影像t同一局部區域灰階值和G(x)相減判斷該局部區域在第二張影像t是否有移動量M _t(x)如下公式：M _t(x)=|G _t(x)-G _t-1(x)|而算出來M _t(x)即為第二張影像t的局部區域x的移動量，對每個局部區域x的連續影像做計算，若該區域產生移動量，在統計中以1來計算，反之，若未產生動量則以0來計算，如下判斷式：Pulse_t(x)=1,If M _t(x)>0 T-1 using the first image and the second image area the same local gray value t, and G (x) of the subtraction is determined whether a movement amount of the local area in second image M _t t (x) the following formula: M _t ( x )=| G _t ( x )− G _{t −1} ( x )| and calculate that M _t ( x ) is the amount of movement of the local region x of the second image t, and a continuous image for each local region x Do the calculation. If the area produces the amount of movement, it is calculated as 1 in the statistics. Otherwise, if no momentum is generated, it is calculated as 0. The following judgment formula: Pulse _t ( x )=1, If M _t ( x )>0

Pulse_t(x)=0,If M _t(x)=0再將各個區域x的物件出現情形以迭代的方式做統計如下公式：將連續的影像△t時間內的各個局部區域脈衝Pulse_t(x)加總過後即可統計出各個局部區域脈衝的次數，亦即新的脈衝加入統計集合中，將會捨去統計集合中最舊的脈衝。 Pulse _t ( x )=0, If M _t ( x )=0 Then, the occurrence of the objects in each region x is calculated in an iterative manner as follows: By summing the local area pulses Pulse _t ( x ) in the continuous image Δt time, the number of pulses of each local area can be counted, that is, the new pulse is added to the statistical set, which will discard the most in the statistical set. Old pulse.

即可得到每個局部區域位置的物件出現頻率，若出現次數越高，統計結果S_t(x)將會越趨近於1，反之，出現次數越少，統計結果S_t(x)將會越趨近於0。 The frequency of occurrence of objects in each local area position can be obtained. If the number of occurrences is higher, the statistical result S _t (x) will be closer to 1, and vice versa, the less the number of occurrences, the statistical result S _t (x) will be The closer to 0.

本發明之另一目的在於若該區域有較高的區域物件出現移動程度，卻不是真正有移動物件在該區域，將可以提高該區域的背景更新速度，快速地消除非背景模組所偵測到的物件。 Another object of the present invention is to improve the background update speed of the area if the object has a higher degree of movement in the area, but the moving object is not in the area, and the non-background module is quickly detected. The object that arrived.

本發明之另一目的在於若有真正的移動物體出現在該區域，變成不動時，能夠以較低的區域物件出現移動程度，減緩該區域的背景更新速度，緩慢地消除背景模組所偵測到的物件，提升背景模組建立的可靠度。 Another object of the present invention is to slowly move the background update speed of the area, and slowly eliminate the background module detection if a real moving object appears in the area and becomes immobile. The objects that are arrived at, improve the reliability of the background module establishment.

本發明之再一目的在於若有遺留物件出現在該區域，能夠以較低的區域物件出現移動程度，減緩該區域的背景更新速度，緩慢地消除背景模組所偵測到的物件，提升背景模組建立的可靠度。 A further object of the present invention is that if a remnant object appears in the area, the degree of movement of the object in a lower area can be reduced, the background update speed of the area can be slowed down, the objects detected by the background module can be slowly eliminated, and the background can be improved. The reliability of the module establishment.

10、11、12、13、14、15‧‧‧步驟 10, 11, 12, 13, 14, 15‧ ‧ steps

101‧‧‧影像集合 101‧‧‧Image collection

101a‧‧‧第一影像 101a‧‧‧ first image

110a(1,0)‧‧‧第一影像座標(1,0)局部區域 110a (1,0)‧‧‧First image coordinates (1,0) local area

110a(8,0)‧‧‧第一影像座標(1,0)局部區域 110a (8,0)‧‧‧First image coordinates (1,0) local area

101b‧‧‧第二影像 101b‧‧‧second image

110b(1,0)‧‧‧第二影像座標(1,0)局部區域 110b (1,0)‧‧‧Second image coordinate (1,0) local area

110b(8,0)‧‧‧第二影像座標(1,0)局部區域 110b (8,0)‧‧‧Second image coordinate (1,0) partial area

110‧‧‧迭代影像 110‧‧‧ Iterative imagery

110(1,0)‧‧‧迭代影像座標(1,0)局部區域 110(1,0)‧‧‧ Iterative image coordinates (1,0) local area

110(1,0)S‧‧‧迭代影像座標(1,0)局部區域堆疊 110(1,0)S‧‧‧ Iterative image coordinates (1,0) local area stacking

110(1,0)F‧‧‧迭代影像座標(1,0)局部區域堆疊中第一位置 110(1,0)F‧‧‧ Iterative image coordinates (1,0) first position in the local area stack

110(1,0)L‧‧‧迭代影像座標(1,0)局部區域堆疊中最後位置 110 (1,0)L‧‧‧ Iterative image coordinates (1,0) the last position in the local area stack

110(8,0)‧‧‧迭代影像座標(8,0)局部區域 110(8,0)‧‧‧ Iterative image coordinates (8,0) local area

110(8,0)S‧‧‧迭代影像座標(8,0)局部區域堆疊 110(8,0)S‧‧‧ Iterative image coordinates (8,0) local area stacking

110(8,0)F‧‧‧迭代影像座標(8,0)局部區域堆疊中第一位置 110(8,0)F‧‧‧ Iterative image coordinates (8,0) first position in the local area stack

110(8,0)L‧‧‧迭代影像座標(8,0)局部區域堆疊中最後位置 110(8,0)L‧‧‧ Iterative image coordinates (8,0) last position in the local area stack

第1圖，係本發明提出第一較佳實施例之部分示意圖。 BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a partial schematic view of a first preferred embodiment of the present invention.

第2圖，係本發明提出第一較佳實施例流程圖，為一種局部背景更新的方法。 2 is a flow chart of a first preferred embodiment of the present invention, which is a method for local background update.

第3圖，係本發明提出第一較佳實施例之部分示意圖。 Figure 3 is a partial schematic view of the first preferred embodiment of the present invention.

第4圖，係本發明提出第一較佳實施例之部分示意圖。 Figure 4 is a partial schematic view of the first preferred embodiment of the present invention.

由於本發明係揭露一種局部背景更新的方法，其中所利用的一般監視系統概念及邏輯運算，已為相關技術領域具有通常知識者所能明瞭，故以下文中之說明，不再作完整描述，同時，以下文中所對照之圖式，係表達與本發明特徵有關之示意，並未亦不需要依據實際情形完整繪製，合先敘明。 Since the present invention discloses a method for local background update, the general monitoring system concept and logical operation utilized are well known to those skilled in the relevant art, and therefore, the description below will not be completely described, and The drawings referred to in the following text are indicative of the features relating to the features of the present invention and are not required to be completely drawn according to the actual situation.

請參見第2圖，係本發明提出第一較佳實施例之流程圖，為一種局部背景更新的方法，其方法將包含： Referring to FIG. 2, a flow chart of a first preferred embodiment of the present invention is a method for local background update, and the method thereof will include:

步驟10，提供影像集合101(例如由監視器所攝得的監視影像)。影像集合101至少包括具有時間序列關係的第一影像101a與第二影像101b為可見之移動物件的影像。 In step 10, an image collection 101 (such as a surveillance image captured by a monitor) is provided. The image set 101 includes at least an image of the moving object that the first image 101a and the second image 101b have a time-series relationship.

步驟11，請參照第1圖，依照影像集合101的影像大小，建立出迭代影像110，並且切分數個局部區域，如座標(1,0)局部區域稱之110(1,0)，座標(8,0)局部區域稱之110(8,0)儲存後面計算個別局部區域統計結果S_t(x)，再以區域物件出現移動程度大小的數值來決定任何區域做局部背景更新速度，每個區域的較佳尺寸可依照監視器所照的畫面遠近做調整，在此預設一般監視器正常的距離安裝，每個區域以初始尺寸為4像素(pixel)*4像素。 Step 11, please refer to FIG. 1 , according to the image size of the image set 101, the iterative image 110 is created, and the partial regions are segmented, for example, the local region of the coordinate (1, 0) is called 110 (1, 0), coordinates ( 8,0) The local area is called 110(8,0), and the individual local area statistical result S _t ( x ) is calculated, and then the value of the degree of movement of the area object is used to determine the local background update speed of each area. The preferred size of the area can be adjusted according to the distance of the picture taken by the monitor. Here, the normal monitor is installed at a normal distance, and each area has an initial size of 4 pixels (pixel) * 4 pixels.

步驟12，利用時間序列關係第二影像101b與第一影像101a中的各個局部區域中的灰階值差異，來判斷該局部區域是否有產生物件出現移動情形。請參照第3圖，第二影像101b中的局部區域101b(1,0)與第一影像101a中的局部區域101a(1,0)兩區域灰階值相減套入公式為：M _101b(1,0)=|G _101b(1,0)-G _101a(1,0)|所以該局部區域的移動量為M _101b(1,0)；第二影像101b中的局部區域101b(8,0)與第一影像101a中的局部區域101a(8,0)兩區域灰階值相減套入公式為：M _101b(8,0)=|G _101b(8,0)-G _101a(8,0)|所以該局部區域的移動量為M _101b(8,0)。 Step 12: Using the time series relationship between the second image 101b and the grayscale value difference in each local region in the first image 101a, it is determined whether the local region has a moving condition of the generated object. Referring to FIG. 3, the partial region 101b (1, 0) in the second image 101b and the partial region 101a (1, 0) in the first image 101a are subtracted from each other by the grayscale value: M _101b ( 1,0)=| G _101b (1,0)- G _101a (1,0)|so the amount of movement of the local area is M _101b (1,0); the partial area 101b of the second image 101b (8, 0) Subtracting the grayscale value between the two regions of the partial region 101a (8, 0) in the first image 101a, the formula is: M _101b (8,0)=| G _101b (8,0)- G _101a (8 , 0)| So the amount of movement of this local area is M _101b (8, 0).

步驟13，利用步驟12各區域位置所產生的移動量做脈衝轉換，若該區域有產生物件出現移動情形，在統計中以1來計算，反之，若未產生物件出現移動情形則以0來計算，以M _101b(1,0)與M _101b(8,0)套入判斷式結果為：Pulse_101b(1,0)=1,M _101b(1,0)>0 In step 13, the amount of movement generated by the position of each region in step 12 is used for pulse conversion. If there is a moving condition of the object in the region, the calculation is performed by 1 in the statistics, and vice versa, if the movement of the object is not generated, the calculation is performed with 0. The result of nesting M _101b (1,0) and M _101b (8,0) is: Pulse _{101 b} (1,0)=1, M _{101 b} (1,0)>0

Pulse_101b(8,0)=0,M _101b(8,0)=0 Pulse _{101 b} (8,0)=0, M _{101 b} (8,0)=0

步驟14，將步驟13各區域所轉換的脈衝依照對應的局部區域依序丟入堆疊做迭代統計如下公式：針對堆疊連續統計△t張幀做統計，將最新的一筆脈衝丟入統計，並且捨去最舊的一筆脈衝，即可計算出在△t張幀內有多少次的移動物體出現S _t(x)，再將此迭代數值存入迭代影像110中。請參照第4圖及第1圖，將步驟13，Pulse_101b(1,0)丟入堆疊110(1,0)S中的第一個位置110(1,0)F，並且捨去110(1,0)L的脈衝值，即可計算出在△t張幀內有多少次的移動物體出現，再將此迭代數值S₁₁₀(1,0)存入迭代影像110中的局部區域110(1,0)；將Pulse_101b(8,0)，丟入堆疊110(8,0)S中的第一個位置110(8,0)F，並且捨去110(8,0)L的脈衝值，即可計算出在△t張幀內有多少次的移動物體出現，再將此迭代數值S₁₁₀(8,0)存入迭代影像110中的局部區域110(8,0)。 In step 14, the pulse converted in each area of step 13 is sequentially thrown into the stack according to the corresponding local area, and the following formula is calculated: For the stack continuous statistics △ t frames to do statistics, the latest pulse is thrown into the statistics, and the oldest pulse is discarded, you can calculate how many times in the Δt frame, the moving object appears S _t ( x Then, the iteration value is stored in the iterative image 110. Referring to FIG. 4 and FIG. 1 , in step 13, Pulse _{101 b} (1, 0) is dropped into the first position 110 (1, 0) F in the stack 110 (1, 0) S, and 110 is discarded. With the pulse value of (1,0)L, it can be calculated how many times the moving object appears in the Δt frame, and then the iteration value S ₁₁₀ (1, 0) is stored in the partial region 110 in the iterative image 110. (1,0); Drop Pulse _{101 b} (8,0) into the first position 110(8,0)F in stack 110(8,0)S, and discard 110(8,0)L The pulse value can be used to calculate how many times the moving object appears in the Δt frame, and then the iteration value S ₁₁₀ (8, 0) is stored in the partial region 110 (8, 0) in the iterative image 110.

步驟15，請參照第1圖，即可利用每個局部區域110的迭代情形來改變背景更新模組的各個局部區域的更新速度，如背景模組GMM(Gaussian mixture model)中的α值可以依照S_t(x)做適度的調整來改善每個局部區域的更新速度，或是在美國專利8,009,918中所揭露的背景更新模組的方法中，所利用到的Pr值做更新也可以利用S_t(x)做適度的調整改善每個局部區域的更新速度。 Step 15, please refer to FIG. 1 , that is, the iteration situation of each local area 110 can be used to change the update speed of each local area of the background update module, for example, the alpha value in the background module GMM can be S _t ( x ) makes a modest adjustment to improve the update speed of each local area, or in the method of the background update module disclosed in U.S. Patent No. 8,009,918, the Pr value used can be updated to use S _t ( x ) Make moderate adjustments to improve the update speed of each local area.

以上所述僅為本發明之較佳實施例，並非用以限定本發明之權利範圍；同時以上的描述，對於熟知本技術領域之專門人士應可明瞭及實施，因此其他未脫離本發明所揭示之精神下所完成的等效改變或修飾，均應包含在申請專利範圍中。 The above description is only the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. The above description should be understood and implemented by those skilled in the art, and thus the other disclosures are not disclosed. Equivalent changes or modifications made in the spirit of the invention shall be included in the scope of the patent application.

110‧‧‧迭代影像 110‧‧‧ Iterative imagery

110(1,0)‧‧‧座標(1,0)局部區域 110(1,0)‧‧‧partial (1,0) partial area

110(8,0)‧‧‧座標(8,0)局部區域 110(8,0)‧‧‧partial (8,0) partial area

Claims

一種局部背景更新方法，包含：提供一連續影像集合，影像集合至少包含有時間序列關係的第一影像與第二影像，且第一影像早於第二影像，從連續影像集合中，依序的比較第二影像與第一影像中物體的位置差異，觀察各個局部區域物件出現移動情形，並且將各個區域物件出現移動情形以短暫的時間內以迭代的方式做統計，進而得知該區域物件出現移動程度大小，即可調整該區域的背景更新速度，其中該迭代統計的方式是：其中，Pulse(Mi(X))為對應各個局部區域物件出現移動情形，其中當一局部區域物件出現移動情形，對應之Pulse(Mi(X))為1，以及當一局部區域物件未出現移動情形，對應之Pulse(Mi(X))為0。 A partial background update method includes: providing a continuous image set, the image set includes at least a first image and a second image in a time series relationship, and the first image is earlier than the second image, from the continuous image set, sequentially Comparing the difference between the position of the second image and the object in the first image, observing the movement of the objects in each local area, and performing the iteration in a short period of time by moving the movement of each area object, thereby knowing that the object appears in the area. The degree of movement can be adjusted to adjust the background update speed of the area. The iterative statistics are as follows: Among them, Pulse(Mi(X)) is a movement situation corresponding to each local area object, wherein when a local area object moves, the corresponding Pulse(Mi(X)) is 1, and when a local area object does not appear to move In the case, the corresponding Pulse (Mi(X)) is 0.

如申請專利範圍第1項之局部背景更新方法，其中該背景更新方法為局部性的。 The partial background updating method of claim 1, wherein the background updating method is local.

如申請專利範圍第2項之局部背景更新方法，其中該局部區域的計算單位為影像區塊。 The partial background updating method of claim 2, wherein the calculation unit of the local area is an image block.

如申請專利範圍第2項之局部背景更新方法，其中該局部區域的尺寸為4像素*4像素。 The partial background updating method of claim 2, wherein the partial area has a size of 4 pixels * 4 pixels.