LU101981B1

LU101981B1 - Traffic video background modeling method and system

Info

Publication number: LU101981B1
Application number: LU101981A
Authority: LU
Inventors: Yong Qi; Qia Wang
Original assignee: Univ Nanjing Sci & Tech
Priority date: 2020-04-12
Filing date: 2020-07-13
Publication date: 2021-05-18
Also published as: CN111539273A; CN111539273B; WO2021208275A1

Abstract

The present invention discloses a traffic video background modeling method and system. The method comprises the following steps: 1) graying original video frames; 2) extracting a foreground region of adjacent frames by using an inter-frame difference method and determining a background region; 3) determining a pixel value of each position in the background region by using a statistical histogram method; 4) cycling the first three steps in an N-frame video image sequence to reconstruct a background image; and 5) updating the background by using an update strategy of "first out and end in". The system comprises the following modules: a video capture module configured to provide traffic video stream information, a method integration module configured to package a background modeling method, a calculation module configured to execute program functions and process data, a storage module configured to store an application program, source data and processing results, and a display module configured to display input and output image information. The present invention is easy to implement, can be used in an intelligent surveillance system to extract a background image with high cleanliness, and effectively solves the problem of incomplete extraction of traffic backgrounds with slow vehicles.

Description

Description LU101981

TRAFFIC VIDEO BACKGROUND MODELING METHOD AND SYSTEM Field of the Invention The present invention relates to the field of intelligent video analysis technology, and specifically to a traffic video background modeling method and system. Background of the Invention In recent years, rapid development of information technology and sophisticated management of traffic surveillance have jointly ensured the normal operation of road traffic. Video surveillance technology has been researched and applied to a greater and greater extent especially in transportation systems, playing an important role in promoting the development of transportation intelligence. The widespread application of video surveillance technology is accompanied by a large amount of surveillance video data. Background modeling is performed using a video sequence, then a moving target is detected, and subsequent tasks such as vehicle counting, target identification and tracking are undertaken thereby. Therefore, background modeling is a very important research topic in the field of intelligent transportation.

Traffic videos captured by a fixed camera can be used for target vehicle detection. Common methods include an inter-frame difference method, an optical flow method and a background difference method. However, the inter-frame difference method is greatly affected by the speed and the calculation of the optical flow method is relatively complicated, so the methods have obvious defects and hardly meet the requirements of detection systems. The background difference method is simple and easy to use. The most complete feature information and the most suitable target contour can be extracted by referring to a known background. Whether the background difference method can obtain good results depends on whether the background modeling method can extract high-quality background images, which is crucial to the research on the background modeling method.

There is a problem of “bootstrapping” in background modeling. Bootstrapping means pulling up by one's own shoelaces, which is a metaphor for an unrealizable practice. In 1

Description LU101981 background modeling, because there are moving objects such as vehicles and pedestrians at almost every moment in the traffic scene, it is difficult for people to obtain, specially for background training, a “clean” traffic background frame that does not include foreground objects during normal times of each intersection. Accordingly, since foreground targets always appear in the traffic video during background modeling, what needs to be considered first is how to avoid the interference of foreground targets in the Bootstrapping scene, so as to obtain true and complete background image information.

In view of the above-mentioned problem analysis and the characteristics of urban traffic, a new and better background modeling solution is needed, in order to solve the problem that the detection of foreground targets is inaccurate because it is difficult to extract the traffic background from an urban road traffic video. Summary of the Invention The objective of the present invention is to provide a traffic video background modeling method and system, to solve the problem that the detection of foreground targets is inaccurate because it is difficult to extract the traffic background from an urban road traffic video.

The technical solution to achieve the objective of the present invention is as follows: A traffic video background modeling method includes the following steps: step 1: graying original video image frames; step 2: extracting a foreground region of adjacent frames by using an inter-frame difference method and determining a background region; step 3: determining a pixel value of each position in the background region by using a statistical histogram method; step 4: cycling the first three steps in an N-frame video image sequence to reconstruct a background image; and step 5: updating the background by using an update strategy of “first out and end in”.

Further, the step 1 of graying original video image frames is as follows: assuming the RGB composition of a coordinate {x.v} pixel to be (R, G, B), assigning weights to all color 2

Description LU101981 channels, weighting and summing, and then obtaining a gray value V from equation D, wherein the value range of each color channel is [0, 255], so the value range of the gray value V is also [0, 255] F = 0.308 + 0.596 + 0,115 D.

Further, step 2 includes the following sub-steps: step 2.1: for a video frame image sequence F; , #7, , Fy, sequentially differencing every two adjacent images by using an inter-frame difference method to distinguish a background region and a foreground region in each frame of image except the first frame, assuming F,—+ and F, to be two adjacent frames of video images {3 < X = N}, and calculating the difference according to the gray values to obtain a difference image D by equation @ Die as ag) = Planet” Pietro @ step 2.2: selecting an appropriate threshold T for binarizing the difference image D to obtain a binarized image B by equation (3), wherein the point having a gray value of 255 is a background point, and the point having a gray value of 0 is a moving point, i.e., a foreground point (255, if De-vntn ET SL Tig ; Ff Bogota CT ® step 2.3: extracting a video foreground target according to the threshold T, performing an opening operation and a closing operation on the extracted foreground image by using morphology, and performing multiple combined operations of expansion and erosion to reduce the influence of noise, so that the overall contour of a foreground moving target is clearer; and step 2.4: for each foreground moving target region, calculating its enclosing rectangle as a detection box to mark a foreground region M, wherein once the foreground region M is extracted, the background region B is also determined accordingly.

Further, step 3 includes the following sub-steps: 3

Description LU101981 step 3.1: marking all the foreground regions M as -1 to distinguish the gray values in the interval of [0, 255]; step 3.2: establishing a corresponding gray histogram that is not related to the foreground but related to the background for each position {x,¥} in the background region B, counting the occurrence frequency of the gray value of each pixel, and selecting the pixel value p with the most occurrences as the pixel value of the background image at the same coordinate position {x,w), wherein the selection strategy of the pixel value is represented by equation © A { May} = Life yin NM, @ PE ta v= M ax{Hist, y plex vie B, where Hist, {p]= K{x.v.p) + +.if Fay) =r.pe [0,255 © where in equation ©, Æ{x,v.p} represents the number of occurrences when the gray value of a pixel at the image ix.¥} is », A, 1x 5} =p represents that the pixel value of the image #, at {x.¥} is », and Mist, represents a histogram using the gray value © of the pixel as a statistical basis at the coordinates {x, +} and step 3.3: after the pixel value is selected, the foreground region and the background region together forming a background image Bg to be optimized.

Further, in the iterative process of step 4 within N frames, as N increases, the background positions in the video sequence that are always covered by foreground targets gradually decrease; once a new background region is detected out by the inter-frame difference, this part of region contained in the foreground region is updated as the background region, then a new statistical histogram is formed until all regions are updated, and a complete and neat background image is finally obtained.

Further, step 5 includes the following sub-steps: step 5.1: in the corresponding time of N frames, selecting a larger value for N only in the first background modeling in order to ensure the integrity of the initial background image 4

Description LU101981 because there is a high probability that part of the background is always covered by the foreground targets, resulting in a “black hole” region in the background; step 5.2: taking the N-frame gray image sequence as a &ateh, calling the first N frames of images {F,, F2; m, Fy} as &atehy, obtaining a gray histogram Hist, of each pixel position {x.v) by statistics on its corresponding gray value sequence Posguenceizad = Ray Fix. y}, … Fy (x, 23, and finally obtaining a background image Egy; step 5.3: when the (N+1)® frame of image is received in the method, inserting Fy, into batok,, removing F therefrom to obtain batch: {F,, F4. … Fr 411, the gray value sequence corresponding to the gray histogram Hist, being Pasquenesixat = Fores Fataate om Fr wi {x,3}}, and finally obtaining a background image Bag; step 5.4: outputting Bg, if Eg. does not have a “black hole” region; otherwise, taking Bg, as an optimization reference object, filling the gray value of Bg; at the corresponding position in the “black hole” region to Fg. to obtain a complete background image; and step 5.5: if the video stream does not end, continuing to generate new background images according to steps 5.3 and 5.4 until the end of the video stream.

The traffic video background modeling method of the present invention integrates the advantages of an inter-frame difference method and a statistical histogram method, and overcomes the problem that the detection of foreground targets is inaccurate because it is difficult to directly extract a background from a traffic video. The characteristics of a rough region of moving targets are first detected by using the inter-frame difference method to remove a moving region, gray value statistics and selection are then performed on the background image by using the statistical histogram method, a high-quality background

Description LU101981 image is finally obtained after multiple optimizations and reconstructions, and the background is updated in real time.

A traffic video background modeling system includes a video capture module configured to provide continuous traffic video stream information, a method integration module configured to package a background modeling method, a calculation module configured to execute program functions and process data, a storage module configured to store an application program, source data and processing results, and a display module configured to display input and output image information.

Further, the video capture module captures a real-time traffic video stream at a vertical viewing angle of 90° by using a camera fixed on a traffic surveillance pole, a built-in graphics processing unit of the camera processes the captured still images and video image data, and the processed data stream information is stored in the storage module, displayed on the display module, and transmitted to the method integration module as input information; Further, the method integration module is a package of the traffic video background modeling method with an interface reserved to form a black box, and its input is image data in a correct format; Further, the calculation module, as a core calculation unit, implements program calculation and data processing by executing the software program stored in the storage module and calling the image data stored in the storage module; Further, the storage module is configured to store the software program of the background modeling method, the source image data transmitted by the video capture module, and the background image results processed by the calculation module; Further, the display module is used as an image presentation carrier to display input video image information and output background image information.

Compared with the prior art, the advantages of the present invention are: 1) the present invention uses the inter-frame difference to effectively utilize the dynamics of pixels in time series and space, and therefore has higher accuracy; 2) the present invention integrates the advantages of classical methods, and the calculation method is simple and easy to implement, and has good real-time performance; 3) the present invention can accurately capture every 6

Description LU101981 background point, realizes background modeling in as few frames as possible, and has a relatively fast speed; and 4) the present invention still has higher integrity under the traffic scene with slow vehicles. Brief Description of the Drawings FIG. 1 is a flowchart of a traffic video background modeling method according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of a traffic video background modeling system according to an embodiment of the present invention.

FIG. 3 is a comparison diagram of background modeling processes of the present invention and some existing methods under a simulated video.

FIG. 4 is a comparison diagram of integrity change curves of background modeling in the present invention and some existing methods.

FIG. 5 is a comparison diagram of background modeling processes of the present invention and some existing methods under a real video. Detailed Description of the Embodiments In order to understand the objectives, technical solutions and advantages of the present invention more clearly, the content of the present invention will be further described with reference to the accompanying drawings. The specific embodiments described herein are merely used for interpreting the present invention, rather than limiting the present invention.

Embodiments FIG. 1 is a flowchart of a traffic video background modeling method according to an embodiment of the present invention. As shown in FIG. 1, after a video stream is imported, the method steps are as follows: (1) Original video image frames are grayed to reduce the complexity of the method and increase the calculation speed. The specific operation is as follows: The RGB composition of a coordinate {z.v} pixel is assumed to be (R, G, B), all color 7

Description LU101981 channels are assigned with weights, weighted and summed, and then a gray value V is obtained from equation (D. The value range of each color channel is [0, 255], so the value range of the gray value V is also [0, 255].

F = 0308 + 0.596 + 0.115 D The grayed video frame image sequence is denoted as F,.F,. …, F, € IY, N is the total number of frames in the image sequence, # and represent the sizes of each frame of image, i.e., # represents the height of the image, and represents the width of the image. A gray image having a size identical to that of a video frame and having a pixel value of 0 is created as an initial background model for subsequent optimization and update.

(2) An approximate moving region of each frame in the video is extracted as a foreground region by using an inter-frame difference method through the operations of image difference, binarization, mathematical morphological filtering, connectivity analysis, etc., and a background region is determined. The specific operation is as follows: The inter-frame difference method can capture the changes in gray values of two adjacent frames, and define the natures of regions according to the changes. The region composed of points with large gray value changes is denoted as a foreground region 3, and the region composed of points with small gray value changes is denoted as a background region &. Since M is obtained from the difference between the current frame and the previous frame, the part of motion change of the previous frame remains on Af, which causes the detection box to not completely match the actual moving object, and to be slightly larger than the region where the moving object is located. Moreover, the method of the present invention updates & and Æ in the process of continuous iterations to obtain complete background images. Thus, # and £ are not constant, but have a fluctuant relationship.

For the video frame image sequence #,,F;;…,#y, every two adjacent images are sequentially differenced using the inter-frame difference method to distinguish a background 8

Description LU101981 region and a foreground region in each frame of image except the first frame. F,_, and F, are assumed to be two adjacent frames of video images {2 = X < N}, and the difference is calculated according to the gray values to obtain a difference image D by equation 2).

An appropriate threshold T is selected for binarizing the difference image D to obtain a binarized image B by equation ©. The point having a gray value of 255 is a background point, and the point having a gray value of 0 is a moving point, i.e., a foreground point.

Be 243 Ÿ Deen ET © RATS AO, EP Degré = 7 A video foreground target is extracted according to the threshold T, an opening operation and a closing operation are performed on the extracted foreground image by using morphology, and multiple combined operations of expansion and erosion are performed to reduce the influence of noise, so that the overall contour of a foreground moving target is clearer.

For each foreground moving target region, its enclosing rectangle is calculated, and the enclosing rectangle is used as a detection box to mark a foreground region M. Once the foreground region M is extracted, the background region B is also determined accordingly.

(3) The foreground region is marked using a statistical histogram method, a gray value distribution of the image in the background region is obtained, the pixel value of each position in the background region is determined, and the background image is estimated. The specific operation is as follows: All the foreground regions M are marked as -1 to distinguish the gray values in the interval of [0, 255].

A corresponding gray histogram that is not related to the foreground but related to the background is established for each position &x.v} in the background region B, the occurrence frequency of the gray value of each pixel is counted, and the pixel value p with the most occurrences is selected as the pixel value of the background image at the same 9

Description LU101981 coordinate position {x,¥ 1. The selection strategy of the pixel value is represented by equation @.

{ Molar = —1, 4F{x.v} in M, @ CE LB, {x, y} = Max {FF St er fol À éfix, vin 5, Where Hist, fp]= K(x vp} ++. Fay} = n.pe [0255] ® In equation ©, Æ{x,w,r} represents the number of occurrences when the gray value of a pixel at the image {x,v} is @, fi{x.¥} = 5 represents that the pixel value of the image Fp at {x.y} is p, and Fist, represents a histogram using the gray value p of the pixel as a statistical basis at the coordinates {x, #. In equation ©, B,{x,v} uses the pixel value of the pixel (x, y) with the maximum frequency on the N-frame gray-scale video image sequence as a background gray value of the pixel, and the foreground region is marked with -1 in My {x,y} for subsequent update.

After the pixel value is selected, the foreground region and the background region together form a background image Bg to be optimized.

(4) Steps 1, 2, and 3 are cycled in the N-frame video image sequence, the foreground regions are substituted into background regions frame by frame, and a complete and neat background image is finally obtained. The specific operation is as follows: In the iterative process within N frames, as N increases, the background positions in the video sequence that are always covered by foreground targets gradually decrease. Once a new background region is detected out by the inter-frame difference, this part of region contained in the foreground region is updated as the background region, then a new statistical histogram is formed until all regions are updated, and a complete and neat background image is finally obtained.

(5) The integrity of the latest frame of background image is analyzed by using an update strategy of “first out and end in”, and optimization is performed based on the last frame of background image. The specific operation is as follows:

Description LU101981 In the corresponding time of N frames, there is a high probability that part of the background is always covered by the foreground targets, resulting in a “black hole” region in the background. In order to ensure the integrity of the initial background image, a larger value is selected for N only in the first background modeling.

The N-frame gray image sequence is taken as a Satch, and the first N frames of images {FF on Fy} are called &atch,. The gray histogram Hist, of each pixel position {x, xy} is obtained by statistics on its corresponding gray value sequence Proquencelaat 7 iF (a, vi Fifa, vin. Fy (vd), and the background image finally obtained is Bg.

In the method, when the (N+1)® frame of image is received, Fy, is inserted into Batck,, F, is removed to obtain batoh,: {F,. F3. Fragt, the gray value sequence corresponding to the gray histogram Hist, is Posquencerat = Fate Fafagds oo Frau (2:31 and the background image finally obtained is Eg.

If &g, does not have a “black hole” region, it is output; otherwise, taking Eg, as an optimization reference object, the gray value of #34 at the corresponding position in the “black hole” region is filled to Hg, to obtain a complete background image.

And so on, until the end of the video stream.

FIG. 2 is a schematic diagram of a traffic video background modeling system according to an embodiment of the present invention. In the system of the present invention, a video capture module is configured to provide continuous traffic video stream information, a method integration module is configured to package a background modeling method, a calculation module is configured to execute program functions and process data, a storage module is configured to store an application program, source data and processing results, and 11

Description LU101981 a display module is configured to display input and output image information.

As shown in FIG. 2, the video capture module captures a real-time traffic video stream at a vertical viewing angle of 90° by using a camera fixed on a traffic surveillance pole, a built-in graphics processing unit of the camera processes the captured still images and video image data, and the processed data stream information is stored in the storage module, displayed on the display module, and transmitted to the method integration module as input information; The method integration module is a package of the traffic video background modeling method with an interface reserved to form a black box, and its input is image data in a correct format; As a core calculation unit, the calculation module implements program calculation and data processing by executing the software program stored in the storage module and calling the image data stored in the storage module; The storage module is configured to store the software program of the background modeling method, the source image data transmitted by the video capture module, and the background image results processed by the calculation module; The display module is used as an image presentation carrier to display input video image information and output background image information.

The present invention integrates the advantages of the inter-frame difference method and the statistical histogram method, packages the method into a module, is supported by an intelligent surveillance system, and therefore, can overcome the problem that the detection of foreground targets is inaccurate because it is difficult to directly extract a background from a traffic video. This solution specifically innovates in that the present invention integrates the advantages of classic methods, the calculation method is simple and easy to implement, the inter-frame difference method effectively utilizes the dynamics of pixels in time series and space, and the statistical histogram method effectively estimates pixel values, so the present invention has higher accuracy, faster calculation speed and higher background integrity. Whether in a normal traffic scene or a typical traffic scene with slow vehicles, this method can extract background images having higher similarity matching with the real background.

12

Description LU101981 In order to verify that the method of the present invention has better effects than the existing technologies, a simulated traffic video and a real traffic video are used for joint verification. The simulated traffic video is used to verify the reliability of theories and principles of the method, and the real traffic video is used to verify the effectiveness of practical applications of the method.

FIG. 3 is a comparison diagram of background modeling processes of the present invention and some existing methods under a simulated video. The resolution of the simulated traffic video is 590x350 pixels, an urban road as the background is denoted as 85,» and a moving vehicle as the foreground is denoted as Fg... IN order to solve the problem that the existing background modeling method does not work well when the vehicle is driven slowly, the vehicle is defined to travel right at a speed of 2-4 pixels per frame. In this typical scene, the performance advantages and disadvantages of a multi-frame image averaging method, a statistical histogram method, a mixed Gaussian background modeling method and the method of the present invention are compared.

As shown in FIG. 3, line a shows simulated video sequence frames, line b shows a background modeling process of the multi-frame image averaging method, line c shows a background modeling process of the statistical histogram method, line d shows a background modeling process of the mixed Gaussian background modeling method, and line e shows a background modeling process of the method of the present invention. Affected by the slow motion of a vehicle, part of the background is blocked by the vehicle for a long time. Background images extracted by the multi-frame image averaging method are uneven in pixel distribution and have obvious traces of distortion. The statistical histogram method can finally extract a background image that is closer to the actual background, but still some noise remains, and the performance of this method is still not ideal in a complex scene environment. The mixed Gaussian background modeling method uses multiple Gaussian distributions to describe the color presentation law of each pixel, which has high time complexity and will introduce noise. The method of the present invention can extract the most complete and clearest background image at the 47% frame. Compared with the other three methods, the 13

Description LU101981 method of the present invention can still maintain good performance even in a scene of slow vehicle motion.

FIG. 4 is a comparison diagram of integrity change curves of background modeling in the present invention and some existing methods. Theoretically, the most direct way to measure the integrity of a background image is to judge the consistency of the extracted background image Hg and the real background image Hg true at the pixel level. To facilitate comparison, the following equation is defined: \ . Ez vd {{BynFs dr) NEPOR = LEN TRE NBFOR Ea Fæsrue © NBFOR (Non-Background pixels to Foreground Objection pixels Ratio) refers to the ratio of pixels having different pixel values in 8g and Bg... to pixels in the foreground image Fg...» and a smaller NBFOR value indicates fewer unreal background pixels remaining in #g, which means higher integrity.

FIG. 4 shows integrity change processes of background extraction of various methods in an entire simulated video. X axis represents the video frame number, and Y axis represents the NBFOR value. It can be seen that the method of the present invention can extract the background image with the highest integrity by using fewer frames.

FIG. 5 is a comparison diagram of background modeling processes of the present invention and some existing methods under a real video. Data is acquired from a UA-DETRAC dataset, which has a resolution of 960x540 pixels per frame at 25 frames per second. In a real traffic scene, the performance advantages and disadvantages of a multi-frame image averaging method, a statistical histogram method, a mixed Gaussian background modeling method and the method of the present invention are compared.

As shown in FIG. 5, line a shows real video sequence frames, line b shows a background modeling process of the multi-frame image averaging method, line c shows a background modeling process of the statistical histogram method, line d shows a background modeling process of the mixed Gaussian background modeling method, and line e shows a background 14

Description LU101981 modeling process of the method of the present invention. The method of the present invention can obtain a complete background image when N is 15, uses fewer frames, and is faster in time and higher in integrity than other methods.

The embodiments described above are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. The terms and symbols used above are intended to best explain the principles and processes of each embodiment, so that other skilled in the art can understand the embodiments described herein. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the design spirit of the present invention shall fall into the protection scope determined by the claims of the present invention.

Claims

Claims LU101981

1. A traffic video background modeling method, comprising the following steps: step 1: graying original video image frames; step 2: extracting a foreground region of adjacent frames by using an inter-frame difference method and determining a background region; step 3: determining a pixel value of each position in the background region by using a statistical histogram method; step 4: cycling the first three steps in an N-frame video image sequence to reconstruct a background image; and step 5: updating the background by using an update strategy of “first out and end in”.

2. The traffic video background modeling method according to claim 1, wherein the step 1 of graying original video image frames is as follows: assuming the RGB composition of a coordinate {x,¥} pixel to be (R, G, B), assigning weights to all color channels, weighting and summing, and then obtaining a gray value V from equation (D, wherein the value range of each color channel is [0, 255], so the value range of the gray value V is also [0, 255] F = 0,308 + 0596 + 0.118 D.

3. The traffic video background modeling method according to claim 1, wherein step 2 comprises the following sub-steps: step 2.1: for a video frame image sequence fj, &%, «..Fy (N is the total number of frames of a video stream), sequentially differencing every two adjacent images by using an inter-frame difference method, assuming F;-, and F, to be two adjacent frames of video images {2 = K = N}, and calculating the difference according to the gray values to obtain a difference image D by equation @ EN @ step 2.2: selecting a threshold T for binarizing the difference image D to obtain a binarized image B by equation (3), wherein the point having a gray value of 255 is a background point, and the point having a gray value of 0 is a moving point, i.e., a foreground 16

Claims LU101981 point {255, if Die Mae ST ® anna = to, EF Diana) ST step 2.3: extracting a video foreground target according to the threshold T, performing an opening operation and a closing operation on the extracted foreground image by using morphology, and performing multiple combined operations of expansion and erosion to reduce the influence of noise, so that the overall contour of a foreground moving target is clearer; and step 2.4: for each foreground moving target region, calculating its enclosing rectangle as a detection box to mark a foreground region M, wherein if the foreground region M is extracted, the non-M region is the background region B.

4. The traffic video background modeling method according to claim 1, wherein step 3 comprises the following sub-steps: step 3.1: marking all the foreground regions M as -1 to distinguish the gray values in the interval of [0, 255]; step 3.2: establishing a corresponding gray histogram that is not related to the foreground but related to the background for each position {x,¥} in the background region B, counting the occurrence frequency of the gray value of each pixel, and selecting the pixel value p with the most occurrences as the pixel value of the background image at the same coordinate position {x,¥), wherein the selection strategy of the pixel value is represented by equation @ { M {x.y} = Lx, vin dM, Fg =+1_ + Seer FON ee + © - ® (5, [= pi = Mox\Hist | 5} LE {x,y}in Be where Hist, ufr} = KU ++i Far) =p pe [0255] © where in equation ©), Æ{x,w.p} represents the number of occurrences when the gray value of a pixel at the image {,¥} is p, f(x, y} =p represents that the pixel value of the image f, at {x.v} is ©, and Fist, represents a histogram using the gray value # of the 17

Claims LU101981 pixel as a statistical basis at the coordinates {x, w}; and step 3.3: after the pixel value is selected, the foreground region and the background region together forming a background image Bg to be optimized.

5. The traffic video background modeling method according to claim 1, wherein in the iterative process of step 4 within N frames, as N increases, the background positions in the video sequence that are always covered by foreground targets gradually decrease; once a new background region is detected out by the inter-frame difference, this part of region contained in the foreground region is updated as the background region, then a new statistical histogram is formed until all regions are updated, and a complete and neat background image is finally obtained.

6. The traffic video background modeling method according to claim 1, wherein step 5 comprises the following sub-steps: step 5.1: in the corresponding time of N frames, selecting N from a value range of 20 to in the first background modeling, or selecting N from a value range of 10 to 15 in non-first background modeling; step 5.2: taking the N-frame gray image sequence as a #atch, calling the first N frames of images {F,. Fa, Aut as datehy, obtaining a gray histogram H ist, ,. of each pixel position {x.y} by statistics on its corresponding gray value sequence Prsquencetmyt = th Levi Fa 0e vk Fy (=, 53}, and finally obtaining a background image Bg, of aregion without “black holes”; step 5.3: when the (N+1)" frame of image is received in the traffic video background modeling method, inserting Fy+; into &eaich, , removing A, therefrom to obtain Satcha {Fan Frs vo: Fg 543, the gray value sequence corresponding to the gray histogram Hist, being Prequencetest 7 (FageatFs (gd es Fy fev) , and finally obtaining a background image Sg; 18

Claims LU101981 step 5.4: outputting Bg, if Eg. does not have a “black hole” region; otherwise, taking Bgs as an optimization reference object, filling the gray value of &g, at the corresponding position in the “black hole” region to Eg. to obtain a complete background image; and step 5.5: if the video stream does not end, continuing to generate new background images by returning to steps 5.3 and 5.4 until the end of the video stream.

7. A traffic video background modeling system, comprising the following modules: a video capture module configured to provide continuous traffic video stream information, a method integration module configured to package a background modeling method, a calculation module configured to execute program functions and process data, a storage module configured to store an application program, source data and processing results, and a display module configured to display a generated background image.

8. The traffic video background modeling system according to claim 7, wherein the video capture module captures a real-time traffic video stream at a vertical viewing angle of 90° by using a camera fixed on a traffic surveillance pole, a built-in graphics processing unit of the camera processes the captured still images and video image data, and the processed data stream information is stored in the storage module, displayed on the display module, and transmitted to the method integration module as input information; the method integration module is configured to package a traffic video background modeling method with an interface reserved to form a black box, and its input is image data in a correct format; the calculation module implements program calculation and data processing by executing a software program stored in the storage module and calling the image data stored in the storage module; the storage module is configured to store a software program of the traffic video background modeling method, source image data transmitted by the video capture module, and background image results processed by the calculation module; and the display module functions as an image presentation carrier to display input video image information and output background image information.

19