WO2021208275A1

WO2021208275A1 - Traffic video background modelling method and system

Info

Publication number: WO2021208275A1
Application number: PCT/CN2020/101551
Authority: WO
Inventors: 戚湧; 王恰
Original assignee: 南京理工大学
Priority date: 2020-04-12
Filing date: 2020-07-13
Publication date: 2021-10-21
Also published as: CN111539273A; LU101981B1; CN111539273B

Abstract

Disclosed are a traffic video background modelling method and system. The method comprises the following steps: 1) performing graying processing on an original video frame; 2) extracting, by means of an inter-frame difference method, foreground areas of adjacent frames to determine a background area; 3) determining a pixel value of each position in the background area by means of a statistical histogram method; 4) performing, in a loop, the first three steps in a sequence of N frames of video images to reconstruct a background image; and 5) updating the background by using a "first out and last in" updating policy. The system comprises the following modules: a video collection module for providing traffic video stream information; a method integration module for encapsulating a background modelling method; a computing module for executing program functions and processing data; a storage module for storing an application program, source data and processing results; and a display module for displaying input and output image information. The method and system are easily implemented, are applied to an intelligent monitoring system so that a clean background image can be extracted, and can effectively solve the problem of the incomplete extraction of a traffic background when vehicles move slowly.

Description

Method and system for modeling traffic video background

Technical field

The invention relates to the technical field of intelligent video analysis, in particular to a traffic video background modeling method and system.

Background technique

In recent years, rapid development of information technology and sophisticated management of traffic monitoring have jointly ensured the normal operation of road traffic. Video surveillance technology has been more and more researched and applied, especially in the transportation system, which plays an important role in promoting the development of transportation intelligence. With the widespread application of video surveillance technology, a large amount of surveillance video data has also followed. The use of video sequences for background modeling and then detection of moving targets is used to undertake subsequent tasks such as vehicle counting, target recognition and tracking. Therefore, background modeling is a very important research topic in the field of intelligent transportation.

Traffic video captured by a fixed camera can be used for target vehicle detection. Common methods include inter-frame difference method, optical flow method, and background difference method. However, the inter-frame difference is greatly affected by the speed and the calculation of the optical flow method is more complicated. The method has obvious defects and it is difficult to meet the requirements of the detection system. 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 using the known background as a reference. Whether the background difference method can achieve better results depends on whether the background modeling algorithm can extract high-quality background images, and the research on the background modeling algorithm is very important.

There is a problem of "Bootstrapping" in the background modeling. Bootstrapping means that a person lifts himself up with the shoelaces on his feet, which is a metaphor for an unrealizable practice. In background modeling, it means that because there are moving objects such as vehicles and pedestrians at almost every moment in the traffic scene, it is difficult for people to get a "clean" traffic background frame that does not include foreground objects during normal times of each intersection, specifically for background training . Therefore, there are always foreground targets in the traffic video during background modeling. The first thing to consider 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, in order to solve the problem that the urban road traffic video is difficult to directly extract the traffic background, which leads to the inaccurate detection of foreground objects, a new and better background modeling solution needs to be proposed.

Summary of the invention

The object of the present invention is to provide a traffic video background modeling method and system to solve the problem that it is difficult to directly extract the traffic background from the urban road traffic video, which leads to inaccurate detection of foreground objects.

The technical solutions to achieve the purpose of the present invention are as follows:

A traffic video background modeling method includes the following steps:

Step 1: Perform a grayscale operation on the original video image frame;

Step 2: Use the inter-frame difference method to extract the foreground area of adjacent frames and determine the background area;

Step 3: Use the statistical histogram method to determine the pixel value of each position in the background area;

Step 4: Loop the first three steps in the N-frame video image sequence to reconstruct the background image;

Step 5: Use the "first-out and end-in" update strategy to update the background.

Further, the grayscale operation of the original video image frame in the step 1 is as follows: Set the RGB composition of the coordinate (x, y) pixel point as (R, G, B), assign weights to all color channels and calculate the weights And, the gray value V is obtained by formula ①. The value range of each color channel is [0,255], so the value range of the gray value V is also [0,255].

V=0.30R+0.59G+0.11B①

Further, the step 2 includes the following sub-steps:

Step 2.1: For the video frame image sequence F ₁ , F ₂ , ..., F _N , use the inter-frame difference method to perform pairwise difference of adjacent images in sequence, and determine the difference between the background area and the background area in each image except the first frame. Foreground area. Let F _k-1 and F _k be two adjacent frames of video images (2≤K≤N), calculate the difference according to the gray value, and obtain the difference image D, which is obtained by formula ②.

D _(k-1,k)(x,y) =|F _(k-1)(x,y) -F _k(x,y) | ②

Step 2.2: Select a suitable threshold T to perform binarization operation on the difference image D to obtain a binarized image B, which is obtained by formula ③. Among them, the gray value of 255 is the background point, and the gray value of 0 is the moving point, that is, the front scenic spot.

Step 2.3: Extract the video foreground target according to the threshold T, and use morphology to perform opening and closing operations on the extracted foreground image, and reduce the influence of noise through multiple expansion and erosion combined operations, making the overall contour of the foreground moving target clearer.

Step 2.4: For each foreground moving target area, calculate its circumscribed rectangle, and use the circumscribed rectangle as the detection frame to mark the foreground area M. Once the foreground area M is extracted, the background area B is also determined accordingly.

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

Step 3.1: Mark all the foreground area M as -1 to distinguish it from the gray value in the interval [0,255].

Step 3.2: For each position (x, y) in the background area B, establish a corresponding gray histogram that has nothing to do with the foreground and is related to the background, count the occurrence frequency of the gray value of each pixel, and select the one with the most occurrences. The pixel value p is the pixel value at the same coordinate position (x, y) of the background image. The pixel value selection strategy is represented by formula ④.

Among them, Hist _{x, y} [p] = K(x, y, p)++, if F _k (x, y) = p, p ∈ [0, 255] ⑤

In formula ⑤, K(x, y, p) represents the number of times when the pixel gray value is p at the image (x, y), f _k (x, y) = p represents the image f _k in (x, The pixel value at y) is p, Hist _{x, y} represents a histogram based on the pixel gray value p as the statistical basis at the coordinate point (x, y).

Step 3.3: After the pixel value is selected, the foreground area and the background area together form the background image Bg to be optimized.

Further, in the iterative process of step 4 in N frames, as N increases, the background position in the video sequence that is always covered by the foreground target gradually decreases. Once the inter-frame difference detects a new background area, it will replace this part of the foreground area with the background area, and then form a new statistical histogram, until all areas are updated, and finally get a complete and neat background image.

Further, the step 5 includes the following sub-steps:

Step 5.1: Within the corresponding time of N frames, there is a high probability that part of the background will always be covered by the foreground target, resulting in a "black hole" area in the background. In order to ensure the integrity of the initial background image, only in the first background modeling, N takes a larger value.

Step 5.2: Take N frames of gray-scale image sequence as a batch, and call the first N frames of images {F ₁ , F ₂ ,..., F _N } as batch ₁ . The gray-level histogram of each pixel location (x, y) Hist _{x, y} is based on its corresponding gray-level value sequence p _{sequence (x, y)} = {F ₁ (x, y), F ₂ (x, y) ),..., F _N (x, y)} is calculated, and the final background image is Bg1.

Step 5.3: When the method receives the N+1th frame of image, insert F _N+1 into batch ₁ , and remove F ₁ from it. At this time, it is batch ₂ : {F ₂ , F ₃ ,..., F _N+1 }, the gray value sequence p _{sequence (x, y)} corresponding to the gray histogram Hist _{x, y} = {F _{2(x, y)} , F _{3(x, y)} ,..., F _N+1 (x, y)}, the final acquired background image is Bg ₂ .

Step 5.4: If there is no "black hole" area in _{Bg 2} _{, then output; otherwise, using Bg 1} as the optimized reference object, _{fill the gray value of Bg 1} at the corresponding position on the "black hole" area on Bg ₂ , and get the complete Background image.

Step 5.5: If the video stream is not over, follow steps 5.3 and 5.4 to continue to generate a new background image until the video stream ends.

The traffic video background modeling method of the present invention combines the advantages of two methods of inter-frame difference and statistical histogram, and overcomes the problem that it is difficult to directly extract the background in the traffic video, which leads to inaccurate detection of foreground objects. First, use the inter-frame difference method to detect the characteristics of the general area of the moving target, eliminate the moving area, and then use the statistical histogram to count and select the gray value of the background image. After multiple optimization and reconstruction, a high-quality background image is finally obtained, and real-time Perform background updates.

A traffic video background modeling system, including a video acquisition module for providing continuous traffic video stream information, a method integration module for encapsulating the background modeling method, a calculation module for executing program functions and processing data, and a storage module for storing Application programs, source data and processing results, and the display module is used to display input and output image information.

Further, the video capture module uses a fixed device camera on a traffic monitoring pole to capture a real-time traffic video stream with a vertical viewing angle of 90°. The camera has a built-in graphics processor to process the captured still pictures and video image data, and process Data flow 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 body of the traffic video background modeling method, reserves an interface to form a black box, and the input is image data in the 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 used to store the software program of the background modeling method, the source image data transmitted from the video acquisition module, and the background image result processed by the calculation module;

Further, the display module, as an image presentation carrier, is used 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 sequence and space, and has a higher accuracy; 2) the present invention combines the advantages of classic algorithms and calculates The method is simple, easy to implement, and has good real-time performance; 3) The present invention can accurately capture every background point, realizes background modeling in as few frames as possible, and has a faster speed; 4) The present invention is even at the speed of the vehicle In the slower traffic scene, there is still a high degree of completeness.

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.

Figure 3 is a comparison diagram of the background modeling process of the present invention and some existing methods under the simulation video.

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

Figure 5 is a comparison diagram of the background modeling process of the present invention and some existing methods in real video.

Detailed ways

In order to understand the objectives, technical solutions, and advantages of the embodiments of the present invention more clearly, the content of the present invention will be further described below with reference to the accompanying drawings. The specific embodiments described here are only used to explain the present invention, but not used to limit the present invention.

Example

Fig. 1 is a flowchart of a traffic video background modeling method according to an embodiment of the present invention. As shown in Figure 1, after the video stream is imported, the method steps are as follows:

(1) Perform gray-scale operation on the original video image frame to reduce the complexity of the method and increase the calculation speed. The specific operations are as follows:

Assuming that the RGB composition of the pixel at the coordinates (x, y) is (R, G, B), all the color channels are assigned weights and weighted and summed, then the gray value V is obtained by formula ①. The value range of each color channel is [0,255], so the value range of the gray value V is also [0,255].

V=0.30R+0.59G+0.11B①

Denote the gray-scaled video frame image sequence as F ₁ , F ₂ , ..., F _N ^{∈ I h*w} , where N is the total number of frames in the image sequence, and h and w represent the image size of each frame, that is, h represents The image height, w represents the image width. Create a grayscale image that is consistent with the video frame size and has a pixel value of 0 as the initial background model for future optimization and update.

(2) Using the inter-frame difference method, extract the approximate motion area of each frame in the video as the foreground area through image difference, binarization, mathematical morphology filter processing, connectivity analysis and other operations, and determine the background area. The specific operations are as follows:

The inter-frame difference method can capture the changes in the gray value of two adjacent frames, and define the nature of the area according to the change. The area composed of points with a large gray value change is recorded as the foreground area M, and the gray value change is small The area composed of dots is denoted as the background area B. Since M is obtained from the difference between the current frame and the previous frame at this time, the part of the motion change of the previous frame will remain in M, which causes the detection frame to not completely match the actual moving object, and is slightly larger than the area where the moving object is located. Moreover, the method of the present invention is to update M and B in a continuous iterative process to obtain a complete background image. Therefore, M and B are not fixed, they have a trade-off relationship.

For the video frame image sequence F ₁ , F ₂ , ..., F _N , the inter-frame difference method is used to sequentially perform pairwise difference of adjacent images, and distinguish the background area and the foreground area in each frame of the image except the first frame. Let F _k-1 and F _k be two adjacent frames of video images (2≤K≤N), calculate the difference according to the gray value, and obtain the difference image D, which is obtained by formula ②.

D _(k-1,k)(x,y) =|F _(k-1)(x,y) -F _k(x,y) | ②

Choosing a suitable threshold T to perform binarization operation on the difference image D to obtain the binarization image B, which is obtained by formula ③. Among them, the gray value of 255 is the background point, and the gray value of 0 is the moving point, that is, the front scenic spot.

The video foreground target is extracted according to the threshold T, and the extracted foreground image is opened and closed using morphology, and the influence of noise is reduced through multiple expansion and erosion combined operations, so that the overall contour of the foreground moving target is clearer.

For each foreground moving target area, calculate its circumscribed rectangle, and use the circumscribed rectangle as the detection frame to mark the foreground area M. Once the foreground area M is extracted, the background area B is also determined accordingly.

(3) Use the statistical histogram method to mark the foreground area, obtain the gray value distribution of the image in the background area, determine the pixel value of each position in the background area, and estimate the background image. The specific operations are as follows:

Mark all the foreground areas M as -1 to distinguish them from the gray value in the interval [0,255].

For each position (x, y) in the background area B, create a corresponding gray histogram that is irrelevant to the foreground and related to the background, count the occurrence frequency of the gray value of each pixel, and select the pixel value p with the most occurrences. As the background image, the pixel value at the same coordinate position (x, y). The pixel value selection strategy is represented by formula ④.

In formula ⑤, K(x, y, p) represents the number of times when the pixel gray value is p at the image (x, y), f _k (x, y) = p represents the image f _k in (x, The pixel value at y) is p, Hist _{x, y} represents a histogram based on the pixel gray value p as the statistical basis at the coordinate point (x, y). In formula ④, B _k (x, y) takes the pixel value of the pixel (x, y) with the largest frequency in the N-frame gray-scale video image sequence as the background gray value of the pixel, M _k (x, y) Mark the foreground area with -1 for subsequent updates.

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

(4) Repeat steps 1, 2, and 3 in the N-frame video image sequence, and replace the foreground area into a background area frame by frame, and finally obtain a complete and neat background image. The specific operations are as follows:

In the iterative process within N frames, as N increases, the background position in the video sequence that has been covered by the foreground target gradually decreases. Once the inter-frame difference detects a new background area, it will update the part of the area contained in the foreground area as the background area, and then form a new statistical histogram, until all areas are updated, and finally get a complete and neat background image.

(5) Adopt the update strategy of "first out and end in" to analyze the integrity of the latest frame of background image, and optimize the processing based on the previous frame of background image. The specific operations are as follows:

Within the corresponding time of N frames, there is a high probability that part of the background will always be covered by the foreground target, resulting in a "black hole" area in the background. In order to ensure the integrity of the initial background image, only in the first background modeling, N takes a larger value.

Taking a sequence of N frames of grayscale images as a batch, the first N frames of images {F ₁ , F ₂ ,..., F _N } are called batch ₁ . The gray-level histogram of each pixel location (x, y) Hist _{x, y} is based on its corresponding gray-level value sequence p _{sequence (x, y)} = {F ₁ (x, y), F ₂ (x, y) ),..., F _N (x, y)} is calculated, and the final background image is Bg ₁ .

When the method receives the N+1th frame image, it inserts F _N+1 into batch ₁ , and removes F ₁ from it. At this time, it is batch ₂ : {F ₂ , F ₃ ,..., F _{N+ 1} }, the gray value sequence p _{sequence (x, y)} corresponding to the gray histogram Hist _{x, y} = {F _{2(x, y)} , F _{3(x, y)} ,..., F _N+1 (x, y)}, the final acquired background image is Bg ₂ .

If Bg ₂ "black hole" area does not exist, then the output; otherwise, in order to optimize the reference object Bg _1, Bg ₁ on the "black hole" area fill gradation value at the corresponding position to the Bg _2, to obtain a complete image background.

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, the video acquisition module is used to provide continuous traffic video stream information, the method integration module is used to encapsulate the background modeling method, the calculation module is used to execute program functions and process data, and the storage module is used to store application programs and source data. And processing results, the display module is used to display input and output image information.

As shown in Figure 2, the video capture module uses a fixed-device camera on a traffic monitoring pole to capture a real-time traffic video stream with a vertical viewing angle of 90°. The camera has a built-in graphics processor to process the captured still pictures and video image data, and The processed data flow 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 the package body of the traffic video background modeling method. The interface is reserved to form a black box, and the input is image data in the correct format;

As the core computing 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 used to store the software program of the background modeling method, the source image data from the video acquisition module and the background image result 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 invention integrates the advantages of the two methods of inter-frame difference and statistical histogram, and encapsulates the method into a module, and is supported by an intelligent monitoring system, which can overcome the problem that it is difficult to directly extract the background in the traffic video and the foreground target detection is inaccurate. The innovations in the scheme are specifically: the present invention integrates the advantages of classic algorithms, the calculation method is simple and easy to implement, the inter-frame difference is used to effectively utilize the dynamics of pixels in time sequence and space, and the statistical histogram is used to effectively estimate the pixel value. 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 a background image that is similar to the real background and has a higher degree of matching.

In order to verify that the method of the present invention has a better effect than the prior art, the simulated traffic video and the real traffic video are used for joint verification. Among them, the simulation traffic video is used to verify the reliability of the theoretical principle of the method, and the real traffic video is used to verify the effectiveness of the method in practical application.

Figure 3 is a comparison diagram of the background modeling process of the present invention and some existing methods under the simulation video. The resolution of the simulated traffic video is 590×350 pixels. The city road is recorded as Bg _true as the background, and the moving vehicle is recorded as Fg _true as the foreground. In order to solve the problem that the existing background modeling method does not work well when the vehicle is traveling slowly, the vehicle is defined to travel right at a speed of 2-4 pixels per frame. In this typical scenario, compare the performance pros and cons of the multi-frame image averaging algorithm, statistical histogram algorithm, mixed Gaussian background modeling algorithm and the method of the present invention.

As shown in Figure 3, where a is the simulation video sequence frame, b is the background modeling process of the multi-frame image averaging algorithm, c is the background modeling process of the statistical histogram algorithm, and d is the background modeling process of the mixed Gaussian background modeling algorithm. e is the background modeling process of the method of the present invention. Affected by the slow movement of the vehicle, part of the background is blocked by the vehicle for a long time. The pixel distribution of the background image extracted by the multi-frame image averaging algorithm is uneven, and there are obvious traces of distortion. The statistical histogram algorithm can finally extract a background image that is closer to the actual background, but still some noise remains, and the performance is still not ideal in a complex scene environment. The mixed Gaussian background modeling algorithm uses multiple Gaussian distributions to describe the color presentation law of each pixel, which has high time complexity and introduces noise. The method of the present invention can extract the most complete and clear background image at the 47th frame. Compared with the other three algorithms, it can still maintain good performance even in the scene of slow vehicle movement.

Figure 4 is a comparison diagram of the integrity change curves of the background modeling of 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 Bg and the real background image Bg_true at the pixel level. In order to facilitate comparison, define the formula:

NBFOR (Non-Background pixels to Foreground Objection pixels Ratio) refers to the ratio of the _{number of pixels with different pixel values in Bg and Bg true} to the number of pixels in the foreground image Fg _true . The fewer real background pixels, the higher the integrity.

Figure 4 shows the completeness change process of various algorithms extracting the background in the entire simulation video. The X axis represents the video frame number, and the 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 degree of completeness, and it can be realized by using fewer frames.

Figure 5 is a comparison diagram of the background modeling process of the present invention and some existing methods under real video. The data is taken from the UA-DETRAC dataset, which has a resolution of 960×540 pixels per frame at 25 frames per second. In a real traffic scene, compare the performance pros and cons of the multi-frame image averaging algorithm, statistical histogram algorithm, mixed Gaussian background modeling algorithm and the method of the present invention.

As shown in Figure 5, where a is the real video sequence frame, b is the background modeling process of multi-frame image averaging algorithm, c is the background modeling process of statistical histogram algorithm, and d is the background modeling process of mixed Gaussian background modeling algorithm. e is the background modeling process of the method of the present invention. The method of the present invention can obtain a complete background image when N is set to 15. Compared with other methods, the number of frames used is less, the time is faster, and the degree of completeness is higher.

The above-mentioned embodiments merely describe the preferred embodiments of the present invention, and do not limit the scope of the present invention. The terms and formula symbols used in the text are intended to best explain the principles and processes of the embodiments, so that other skilled in the art can understand the embodiments described herein. Without departing from the design spirit of the present invention, various modifications and improvements made by those skilled in the art to the technical solution of the present invention should fall within the protection scope determined by the claims of the present invention.

Claims

A traffic video background modeling method is characterized in that it includes the following steps:

Step 1: Perform a grayscale operation on the original video image frame;

Step 2: Use the inter-frame difference method to extract the foreground area of adjacent frames and determine the background area;

Step 3: Use the statistical histogram method to determine the pixel value of each position in the background area;

Step 4: Loop the first three steps in the N-frame video image sequence to reconstruct the background image;

Step 5: Use the "first-out and end-in" update strategy to update the background.
The traffic video background modeling method according to claim 1, characterized in that, in the step 1, the gray-scale operation of the original traffic video image frame is as follows: Set the RGB composition of the coordinate (x, y) pixel point as ( R, G, B), assign weights to all color channels and weighted sum, then the gray value V is obtained by formula ①, the value range of each color channel is [0,255], and the value range of the gray value V is [0,255]

V=0.30R+0.59G+0.11B①.
The traffic video background modeling method according to claim 1, wherein said step 2 comprises the following sub-steps:

Step 2.1: For the video frame image sequence F 1 , F 2 , ..., F N , N is the total number of video stream frames; use the inter-frame difference method to sequentially differentiate adjacent images in pairs, set F k-1 and F k is two adjacent frames of video images (2≤K≤N), and the difference is calculated according to the gray value to obtain the difference image D, which is obtained by formula ②;

D (k-1,k)(x,y) =|F (k-1)(x,y) -F k(x,y) | ②

Step 2.2: Select the threshold T to perform binarization operation on the difference image D to obtain the binarized image B, which is obtained by formula ③, where the gray value is 255 as the background point, and the gray value is 0 as the motion Point to the former scenic spot;

Step 2.3: Extract the video foreground target according to the threshold T, and use the morphology to open and close the extracted foreground image, and reduce the influence of noise through the combined operation of expansion and erosion, so that the overall contour of the foreground moving target is clearer;

Step 2.4: Calculate the circumscribed rectangle for each foreground moving target area, and use the circumscribed rectangle as the detection frame to mark the foreground area M. If the foreground area M is extracted, the non-M area is the background area B.
The traffic video background modeling method according to claim 1, wherein said step 3 comprises the following sub-steps:

Step 3.1: Mark all the foreground area M as -1, which is used to distinguish the gray value from the interval [0,255];

Step 3.2: For each position (x, y) in the background area B, establish a corresponding gray histogram that has nothing to do with the foreground and is related to the background, count the occurrence frequency of the gray value of each pixel, and select the one with the most occurrences. The pixel value p is used as the pixel value at the same coordinate position (x, y) of the background image, and the pixel value selection strategy is expressed by formula ④;

Among them, Hist x, y [p] = K(x, y, p)++, if F k (x, y) = p, p ∈ [0, 255] ⑤

In formula ⑤, K(x, y, p) represents the number of times when the pixel gray value is p at the image (x, y), f k (x, y) = p represents the image f k in (x, y) the pixel value is p, Hist x, y represents the histogram based on the pixel gray value p as the statistical basis at the coordinate point (x, y);

Step 3.3: After the pixel value is selected, the foreground area and the background area together form the background image Bg to be optimized.
The traffic video background modeling method according to claim 1, characterized in that, in the iterative process of step 4 in N frames, as N increases, the position of the background that has been covered by the foreground target in the video sequence gradually Decrease, once a new background area is detected by the inter-frame difference, the part of the area contained in the foreground area is updated as the background area, and then a new statistical histogram is formed until all areas are updated, and finally a background image is obtained.
The traffic video background modeling method according to claim 1, wherein said step 5 comprises the following sub-steps:

Step 5.1: In the corresponding time of N frames, the value range of N is between 20 and 30 during the first background modeling, and the value of N is between 10 and 15 during the non-first background modeling;

Step 5.2: Take N frames of gray-scale image sequence as a batch, call the first N frames of images {F 1 , F 2 ,..., F N } as batch 1 ; the gray scale of each pixel position (x, y) The histogram Hist x, y is based on its corresponding gray value sequence p sequence (x, y) = {F 1 (x, y), F 2 (x, y),..., F N (x, y) )} Based on statistics, the final complete background image of the area without "black holes" is Bg 1 ;

Step 5.3: When the traffic video background modeling method receives the N+1th frame image, insert F N+1 into batch 1 , and remove F 1 from it. At this time, it is batch 2 : {F 2 , F 3 ,..., F N+1 }, the gray value sequence p sequence (x, y) corresponding to the gray histogram Hist x, y = {F 2 (x, y), F 3 (x, y), ..., F N+1 (x, y)}, the final acquired background image is Bg 2 ;

Step 5.4: If there is no "black hole" area in Bg 2 , then output; otherwise, using Bg 1 as the optimized reference object, fill the gray value of Bg 1 at the corresponding position on the "black hole" area on Bg 2 , and get the complete Background image

Step 5.5: If the video stream is not over, return to steps 5.3 and 5.4 to continue to generate a new background image until the video stream ends.
A traffic video background modeling system, which is characterized by comprising the following modules: a video acquisition module is used to provide continuous traffic video stream information, a method integration module is used to encapsulate a background modeling method, and a calculation module is used to perform program functions and processing Data, the storage module is used to store application programs, source data and processing results, and the display module is used to display the generated background image.
The traffic video background modeling system according to claim 7, characterized in that:

In the video acquisition module, a camera fixedly installed on a traffic monitoring pole collects a real-time traffic video stream with a vertical viewing angle of 90°, and the camera has a built-in graphics processor to process the captured still pictures and video image data, and the processed data The flow 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 used to encapsulate the traffic video background modeling method, reserve an interface, form a black box, and input image data in the correct format;

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 used to store the software program of the traffic video background modeling method, the source image data transmitted from the video acquisition module, and the background image result processed by the calculation module;

The display module, as an image presentation carrier, is used to display input video image information and output background image information.