CN103425764A - Vehicle matching method based on videos - Google Patents

Abstract

The invention discloses a method for quickly matching a moving vehicle in a video. Firstly, the vehicle monitoring video is obtained, and the moving vehicle in each frame image is segmented by using the background frame difference method; secondly, the vehicle position in each frame is recorded to establish the vehicle position Database; place the search window in adjacent image frames centered on the vehicle position to be matched again; then search for the target vehicle in the search window, and adaptively adjust the window size according to the search results until the vehicle is found, and record the vehicle position; finally Calculate the displacement change and speed change of the searched target, select the optimal target, and complete the matching. The proposal of the present invention makes up for the defect that there is no effective method for vehicle matching after the vehicle detection is completed. Moreover, the matching feature proposed by the inventive method is simple to calculate, has low complexity, and has strong real-time performance. The inventive verification module can improve the matching accuracy. have a broad vision of application.

Description

A video-based vehicle matching method

technical field

The invention relates to the technical field of traffic video detection, is a video-based vehicle matching method, and is mainly used for intelligent traffic systems to extract road traffic information.

Background technique

Video-based vehicle matching refers to the identification of the same vehicle at different locations in a series of image frames. At present, many systems (including intelligent transportation systems) need to extract the traffic flow of the road. Many algorithms have successfully detected moving vehicles, but the detection results can only represent the number of vehicles in a single static image. If you need to count the traffic flow, you need to match each vehicle. The perfection of the vehicle matching method is conducive to the intelligentization of traffic. At the same time, the vehicle's driving trajectory can be recorded through vehicle matching. When unsafe behaviors such as illegal lane changes occur, the method can record and report to the traffic management department to improve road traffic safety.

The traditional vehicle matching method is mainly based on the mean-shift method. This method does not take into account the actual engineering application background, and does not optimize the search range during the search, making the algorithm a large amount of calculation and poor real-time performance; secondly, the mean-shift The size of shift's search window is fixed and cannot be adjusted adaptively, which lacks practicality.

Due to the defects and deficiencies of this traditional matching method, it has been difficult to launch mature traffic flow detection products. At the same time, with the development of intelligent transportation systems, the popularity of video acquisition equipment and the cost of video-based traffic information acquisition hardware have gradually decreased. , video-based vehicle matching technology has broad application prospects.

Contents of the invention

Aiming at the defects or deficiencies of the existing vehicle matching technology, the present invention provides a video-based vehicle matching method. This method can quickly and accurately match the moving vehicles within the video range.

In order to realize above-mentioned task, the present invention takes following technical solution:

A video-based vehicle matching method, characterized in the following steps:

Step 1: Obtain the vehicle monitoring video, and use the background frame difference method to segment the moving vehicle in each frame of image.

The vehicle monitoring video is obtained from the video acquisition device, and the moving vehicle in each frame image is segmented by the background frame difference method.

Step 2: Record the position of the vehicle in each frame of image, and establish a vehicle position database.

Record the position of the vehicle in each frame of image, calculate the minimum rectangular outer contour of the vehicle, find the geometric center position of the smallest rectangular outer contour, use this point as the feature point of the vehicle and store it in the database.

Step 3: Center on the vehicle position to be matched, place a search window in adjacent image frames.

Select the vehicle to be tracked, center the geometric center of the minimum rectangular outline of the vehicle, and place a rectangular search window on the adjacent frame to be tracked, which limits the scope of the vehicle search. The size of the rectangular window is 2*T*V*α, where T represents the actual time interval between adjacent frames, V represents the maximum speed limit of the road, and α represents the mapping relationship between the actual distance and the imaging plane in the scene, that is The ratio of the actual distance to the distance in the video.

Step 4: Search for the target vehicle in the search window, and adaptively adjust the size of the window according to the search results until the vehicle is found, and record the vehicle location.

Within the range limited by the search window, the vehicle is searched from the vehicle feature point database. If no vehicle is found, the window size is automatically increased, and the increased step size is 0.2*T*V*α, until the vehicle is found; when the search window size exceeds the monitoring range, the search stops; or when the search window is greater than T* When 405*α, the search stops. Record all the searched vehicle feature points and store them in a set to be screened.

Step 5: Calculate the displacement change and speed change of the searched target, select the optimal target, and complete the matching.

Calculate the displacement change ΔS and speed change ΔV of the vehicle from the set to be screened, and measure the movement state change of the two according to the square sum of the two D=ΔS ² +ΔV ² . Select D as the optimal target for matching.

The invention is based on a video vehicle matching method, and compared with the prior art, it can match all vehicles within the video range. This method has small computational complexity, short matching time and high accuracy, and has broad application prospects.

Description of drawings

Fig. 1 is the flowchart of the method of the present invention.

FIG. 2 is a flow chart of window adaptive adjustment.

Figure 3 is a schematic diagram of the installation of the video capture device.

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

Detailed ways

The video-based vehicle matching method of the present invention adopts the principle that the collected video is processed into a grayscale image sequence of a certain format and size. Vehicles are distributed at different locations on the image over time. The time interval between adjacent images is relatively small, and the displacement of the vehicle moving within this time interval is also very small. Therefore, the vehicle can be matched by calculating the displacement of the vehicle. Since the traveling time is very short, the direction and magnitude of the vehicle's speed change very little. So speed can also be used as a feature for matching vehicles. In the process of matching, it is found that it is not necessary to search for all vehicles in the image for matching, and those vehicles that are far away can be directly excluded. Considering the optimization algorithm, the present invention proposes a search window as the search range. In the actual matching process, the size and speed of the vehicles are different. If the window is set too large, many interference targets will be searched, which will increase the amount of calculation in the subsequent matching process. If the window is set too small, it is difficult to ensure that the vehicle can be searched, and the search task cannot be completed. How to set the size of the window becomes a difficult problem to be solved. The invention proposes a method for adaptively adjusting the size of the window, which can eliminate most of the interference targets while ensuring that the vehicle can be searched.

Specifically, the following steps are used to achieve:

Step 1: Segment the moving vehicle

As shown in Figure 3, by assuming that the video acquisition device on the traffic road acquires real-time traffic video, after performing background frame difference processing on the video, all driving vehicles are segmented, and the segmented vehicles are binarized, Record the location coordinates of each vehicle covered in the image.

Step Two: Determine Vehicle Location

The main purpose of this step is to determine the geometric center of the vehicle. The first step is to determine the minimum rectangular outer contour of the vehicle. In the first step, the binary image after vehicle segmentation has been obtained. Therefore, it is only necessary to determine the smallest rectangular outer contour of the connected domain in the binary image. First organize the coordinates of the pixels covered by the vehicle into a set {(x ₁ , y ₁ ), (x ₂ , y ₂ ), (x ₃ , y ₃ )......(x _i , y _i ) }, where x _i and y _i respectively represent the coordinate position of the i-th pixel. The outer contour of the smallest rectangle can pass the maximum and minimum values of x and y in the coordinate set. Next, sort the collection to find the minimum x coordinate x _min , maximum x coordinate x _max , minimum y coordinate y _min , and maximum y coordinate y _max , respectively. Then the four corner points (x _min , y _min ), (x _min , y _max ), (x max , y _min ), (x _max , y _max ) _of the outer contour of the minimum rectangle can be determined. The geometric center (x _mid , y _mid ) of the minimum outer contour is determined according to the four corner points, where x _mid =(x _min +x _max )/2, y _mid =(y _min +y _max )/2.

Step 3: Place the search window

Place a search window on the target that needs to be matched to search for the minimum distance target around the vehicle. Place a rectangular search window on adjacent frames centered on the geometric center of the smallest rectangular outer contour. The size of the rectangular window is 2*T*V*α, where T represents the actual time interval between adjacent frames, V represents the maximum speed limit of the road, and α represents the mapping relationship between the actual distance and the imaging plane in the scene, that is, the actual The ratio of the distance to the distance in the video. In the actual scene, the image acquisition device we selected is 30 frames per second, so T=1/108000h, and the maximum speed limit stipulated by the road is 60km/h, and α varies according to the calibration results of the image acquisition device. The present invention selects α=12.3p/m, where p/m represents the number of pixels occupied by 1m in the actual distance on the imaging plane. After rounding the result, the length and width of the rectangular window are 14 pixels respectively.

Step 4: The window is adaptively resized until the vehicle is found.

Suppose the geometric center of the i-th vehicle detected is ( _xi , y _i ), and the vector used to store the geometric center of the vehicle in the search window is expressed as Vector.(x _m , y _m ), Vector.x _m Indicates the abscissa of the stored m-th vehicle, and Vector.y _m indicates the ordinate of the m-th vehicle. Take the geometric center (x ₀ , y ₀ ) of the tracked vehicle as the center, and search within a rectangular window whose length and width are a, b. If there is a moving vehicle in the window, store the geometric center of the vehicle in the vector Vector.(x _m , y _m ). Then the vehicle stored in Vector.(x _m , y _m ) is tracked based on position and velocity. If after the search is completed, the number of vehicles in the search window is 0, then the length and width of the search window are increased with δ and ε as the step size respectively, that is, a=a+δ, b=b+ε, until the vehicle is searched .

Step five, calculate the displacement variation and velocity variation of the searched target, select the optimal target, and complete the matching.

Calculate the displacement change ΔS and the speed change ΔV of the vehicle, where ΔS is obtained by calculating the Euclidean distance between the geometric centers of the two vehicles, and the speed change ΔV needs to calculate the difference between the displacement changes of the same vehicle. Then calculate the sum of the squares of the vehicle's displacement change and speed change according to D=ΔS ² +ΔV ² . Find the target with the smallest sum of squares from all the search objects to complete the tracking process.

Claims (6)

1. the vehicle matching process based on video is characterized in that following steps:

Step 1: obtain the vehicle monitoring video, utilize the background frame difference method to be partitioned into the moving vehicle in each two field picture.

Obtain the vehicle monitoring video from video capture device, utilize the background frame difference method to be partitioned into the moving vehicle in each two field picture.

Step 2: record the position of vehicle in each two field picture, set up the vehicle position data storehouse.

At first record the vehicle location in each two field picture, obtain the minimum rectangle outline of vehicle; Then find the geometric center position of minimum rectangle outline, the unique point using this as vehicle; Then deposit the unique point of all vehicles in database again.

Step 3: centered by the vehicle characteristics point of needs coupling, in the adjacent image frame, place search window.

The vehicle of select to need following the trail of, the geometric center of minimum rectangle outline of this vehicle of take is window center, places a rectangular search window on consecutive frame, this search window defines the scope of vehicle search.The length of side of this rectangular window is T*V* α, and wherein T means the time interval between consecutive frame, and V means the maximum limit speed per hour of this road, and α means the mapping relations of actual range and imaging plane in scene, the i.e. ratio of actual range and video middle distance.

Step 4: search for target vehicle in search window, and adjust window size until search vehicle, the registration of vehicle position according to the Search Results self-adaptation.

In the scope limited at search window, search vehicle from the set of vehicle characteristics point data.If do not search vehicle, automatically increase window size, the step-length of increase is 0.2*T*V* α, until search vehicle; When search window exceeds monitoring range, search stops; Perhaps, when searching plain window and be greater than T*405* α, search stops.Record all vehicle characteristics points that search, deposit in a set to be screened.

Step 5: calculate the displacement variable and the velocity variable that search target, select optimal objective, complete coupling.

Calculate displacement variable Δ S and the velocity variable Δ V of vehicle from set to be screened, according to both quadratic sum D=Δ S ²+ Δ V ²Weigh both motion state variable quantities, then choose D and mated as optimal objective.

2. according to the described vehicle fast matching method based on video of claim 1, it is characterized in that: the geometric center that the vehicle characteristics for coupling in step 2 is the vehicle minimum outline.

3. according to the described vehicle fast matching method based on video of claim 1, it is characterized in that: the window for search in step 3 for rectangle, the initial size of rectangular window is T*V* α, wherein T means the real time spacing between consecutive frame, V means the maximum limit speed per hour of this road, α means the mapping relations of actual range and imaging plane in scene, i.e. actual range and ratio at the video middle distance.

4. according to the described vehicle fast matching method based on video of claim 1, it is characterized in that: the size of the search window in step 4 is self-adaptation adjustment.When not searching vehicle, the step-length that window size increases is 0.2*T*V* α, and wherein the increase of window has two upper limits: (1). exceed the detection coverage; (2) exceed T*405* α.

5. according to the described vehicle fast matching method based on video of claim 1, it is characterized in that: the displacement variable that the velocity variable computing method in step 5 are the vehicle front cross frame poor: Δ V=Δ S ₁-Δ S ₂, Δ S wherein ₁The displacement variable that means vehicle minimum rectangle outline geometric center previous frame and real-time frame, Δ S ₂Mean the displacement variable that the vehicle previous frame calculates.

6. according to the described vehicle fast matching method based on video of claim 1, it is characterized in that: displacement variable and the quadratic sum of velocity variable, i.e. D=Δ S that the index of weighing the state of motion of vehicle variable quantity in step 5 is vehicle minimum rectangle profile geometric center ²+ Δ V ².

Images