WO2021003823A1 - Video frame image analysis-based vehicle illegal parking detection method and apparatus - Google Patents

Abstract

A video frame image analysis-based vehicle illegal parking detection method and apparatus. The method comprises: detecting whether there is a vehicle in a video frame image corresponding to a target no-parking area (101); if a vehicle is detected, extracting the vehicle information of the vehicle from the video frame image, and determining whether there is vehicle information in the previous video frame image corresponding to the video frame image in a preset vehicle information list, the preset vehicle information list storing the frame numbers of video frame images and corresponding vehicle information (102); if yes, calculating the similarity between the vehicle information in the previous video frame image and the vehicle information in the video frame image (103); and if the similarity is less than a preset threshold, determining that the vehicle in the video frame image is parked illegally (104). The method is suitable for the detection of illegal parking of vehicles, can improve the recognition rate of illegally parked vehicles on lanes, and can assist relevant departments to find illegally parked vehicles in time.

Description

Vehicle parking violation detection method and device based on video frame picture analysis Technical field

This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on July 11, 2019, the application number is 201910624944.9, and the application name is "Method and device for detecting vehicle parking violation based on video frame picture analysis". The reference is incorporated in the application.

Background technique

With the continuous development of society and economy, the number of vehicles in the city has grown rapidly, and the contradiction between parking demand and the supply of parking lots has become increasingly prominent. Illegal parking has become a chronic disease of the city, which has a serious impact on the overall traffic environment and pedestrian safety of the city. With the development of video detection technology and computer vision technology, parking violation detection has received more and more attention.

At present, vehicles are usually monitored by monitoring equipment installed on each road. Once illegally parked vehicles are found, they will promptly contact relevant departments for handling. However, some roads often lack monitoring equipment and cannot detect illegally parked vehicles in time. Timely handling has caused many motor vehicles to park on prohibited roads at will, causing hidden traffic hazards. At the same time, the efficiency of illegal parking of vehicles through monitoring equipment is low.

Summary of the invention

The present application provides a vehicle parking violation detection method and device based on video frame picture analysis, which is mainly capable of improving the recognition rate of parking violation vehicles on the lane and assisting relevant departments to find the parking violation vehicles in time.

According to the first aspect of this application, a vehicle parking violation detection method based on video frame picture analysis is provided, including:

Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

When a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is the vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, so The preset vehicle information list stores the frame number of the video frame picture and the corresponding vehicle information;

If the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, the vehicle information of the previous video frame picture is calculated to be similar to the vehicle information of the video frame picture degree;

When the similarity is less than a preset threshold, it is determined that the vehicle in the video frame picture has violated a stop.

According to a second aspect of the present application, there is provided a vehicle parking violation detection device based on video frame picture analysis, including:

The detection unit is used to detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

The extraction unit is configured to extract vehicle information of the vehicle from the video frame picture when a vehicle is detected, and determine whether there is a previous video frame picture corresponding to the video frame picture in the preset vehicle information list The vehicle information of the video frame picture and the corresponding vehicle information are stored in the preset vehicle information list;

The calculation unit is configured to calculate the vehicle information of the previous video frame picture and the video frame picture if the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list The similarity of the vehicle information;

The determining unit is configured to determine that the vehicle in the video frame picture illegally stops when the similarity is less than a preset threshold.

According to a third aspect of the present application, there is provided a computer non-volatile readable storage medium, on which computer readable instructions are stored, and when the computer readable instructions are executed by a processor, the following steps are implemented:

Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

When a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is the vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, so The preset vehicle information list stores the frame number of the video frame picture and the corresponding vehicle information;

If the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, the vehicle information of the previous video frame picture is calculated to be similar to the vehicle information of the video frame picture degree;

When the similarity is less than a preset threshold, it is determined that the vehicle in the video frame picture has violated a stop.

According to a fourth aspect of the present application, there is provided a computer device including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions Implement the following steps:

Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

When a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is the vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, so The preset vehicle information list stores the frame number of the video frame picture and the corresponding vehicle information;

If the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, the vehicle information of the previous video frame picture is calculated to be similar to the vehicle information of the video frame picture degree;

When the similarity is less than a preset threshold, it is determined that the vehicle in the video frame picture has violated a stop.

This application provides a vehicle parking violation detection method and device based on video frame picture analysis. Compared with the current way of monitoring vehicles through monitoring equipment installed on various roads, this application can detect the video corresponding to the target no-parking area. Whether there is a vehicle in the frame picture; when a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is a previous video corresponding to the video frame picture in the preset vehicle information list The vehicle information of the frame picture, the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; if there is a previous one corresponding to the video frame picture in the preset vehicle information list The vehicle information of the video frame picture is calculated, and the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; at the same time, when the similarity is less than a preset threshold, it is determined that the The vehicle parking violation in the video frame picture can improve the recognition rate of parking violation vehicles in the lane, and at the same time, it can find the parking violation vehicles in time, assist the relevant departments to find the violation vehicles, improve the efficiency of finding the violation vehicles, thereby reducing Hidden traffic hazards caused by illegally parked vehicles.

Description of the drawings

The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application and do not constitute an improper limitation of the application. In the attached picture:

FIG. 1 shows a flow chart of a method for vehicle parking violation detection based on video frame picture analysis provided by an embodiment of the present application;

FIG. 2 shows a flowchart of another vehicle parking violation detection method based on video frame picture analysis provided by an embodiment of the present application;

FIG. 3 shows a schematic structural diagram of a vehicle parking violation detection device based on video frame picture analysis provided by an embodiment of the present application;

FIG. 4 shows a schematic structural diagram of another vehicle parking violation detection device based on video frame picture analysis provided by an embodiment of the present application;

Fig. 5 shows a schematic diagram of the physical structure of a computer device provided by an embodiment of the present application.

Detailed ways

Hereinafter, the application will be described in detail with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.

As in the background art, at present, since deep learning models require a large amount of labeled data to complete the task requirements for different structured predictions, all sample data need to be labeled during model training, that is, a large amount of manual labeling work is required. However, manual The labeling work requires a lot of repetitive labor, the work content is cumbersome, and the efficiency is low. During model training, all sample data are given the same degree of importance, which is not conducive to focusing effective resources on the most difficult to identify sample data, which leads to model training Low efficiency and low prediction accuracy.

In order to solve the above-mentioned problem, an embodiment of the present application provides a vehicle parking violation detection method. As shown in FIG. 1, the method includes:

101. Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area.

Among them, the target no-parking area is the no-parking area specified by the relevant departments. For the embodiment of this application, aerial photography of the road is carried out through the drone camera to obtain the aerial video. The FFmpeg tool is used to set the start and end time of the aerial video. The time interval of the picture, each video frame picture to be detected can be obtained from the aerial video, and further, the preset vehicle detection model is used to perform vehicle detection on the video frame picture to be detected to determine whether there is a vehicle in the video frame picture. , Go to step 102, if it does not exist, go to step 106, which is to detect whether there is a vehicle in the next video frame picture. The preset vehicle detection model can be a preset yolo V3 vehicle detection model or a preset Mask R-CNN vehicle Detection model.

102. Extract the vehicle information of the vehicle from the video frame picture, and determine whether the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list.

Wherein, the preset vehicle information list stores the frame number of the video frame picture and its corresponding vehicle information. The vehicle information includes the feature vector of the vehicle and the bounding box information of the vehicle in the video frame picture. For the implementation of this application For example, input the video frame picture to be detected into the preset vehicle detection model for vehicle detection. If there is a vehicle in the video frame picture, the preset vehicle detection model is used to extract the vehicle information in the video frame picture, for example, the preset The vehicle detection model is the preset yolo V3 vehicle detection model. Input the video frame pictures to be detected into the preset yolo V3 vehicle detection model for vehicle detection. Because the preset yolo V3 vehicle detection model will reduce the output feature map to the input 1/32, usually the input picture is a multiple of 32. Therefore, the video frame picture to be detected is scaled to 256*256, and then the preset yolo V3 vehicle detection model is input for vehicle detection. If yolo V3 vehicle detection is preset When the model detects a vehicle, it outputs the vehicle information of the vehicle, including the feature vector of the vehicle and the bounding box information of the vehicle. For example, the preset yolo V3 vehicle detection model detects that there are two cars in the video frame picture, then the output The bounding box information of the two cars are recorded as M1 (x1, y1, w1, h1), N1 (x2, y2, w2, h2) and the feature vectors m1 and n1 of the two cars, where x and y represents the coordinate information of the center point of the bounding box where the detected vehicle is located, w and h represent the size of the bounding box where the detected vehicle is located, and both m1 and n1 are 1024-dimensional feature vectors.

In addition, the preset vehicle detection model can also be a preset Mask R-CNN vehicle detection model. The video frame pictures to be detected are input to the preset Mask R-CNN vehicle detection model for vehicle detection. Specifically, the preprocessing is first performed The latter video frame picture is input to the full convolutional network to obtain the corresponding vehicle feature picture, and then a predetermined candidate area ROI is set for each point in the vehicle feature picture to obtain multiple candidate area ROIs, and then these candidate area ROIs are sent Enter the RPN network for binary classification and bounding box regression, filter out a part of the candidate area ROI, and perform the ROIAlign operation on the remaining candidate area ROI, and finally classify these candidate area ROIs. If the Mask R-CNN vehicle detection model is preset Detect the vehicle, generate the bounding box and the mask of the vehicle, and output the vehicle information of the vehicle, that is, the size of the bounding box of the vehicle, the center position information, the feature vector of the vehicle, and the mask of the vehicle can show the outline of the vehicle For example, if the preset Mask R-CNN vehicle detection model detects that there are two cars in the video frame picture, the vehicle information of the two cars will be output, which are recorded as M1(x1,y1,w1,h1),N1( x2, y2, w2, h2), the feature vectors m1 and n1 of the two cars and the masks of the two cars, where x and y represent the coordinate information of the center point of the bounding box of the detected vehicle, and w and h represent the boundary of the detected vehicle The size of the frame, m1 and n1 are both 2048-dimensional feature vectors. Since the preset Mask R-CNN vehicle detection model can identify pixel-level areas, the preset Mask R-CNN vehicle detection model can not only identify the video The vehicle in the frame picture can also improve the accuracy of vehicle recognition.

Further, if a vehicle is detected using a preset vehicle detection model, the vehicle information of the vehicle is extracted, and the frame number of the video frame picture is obtained, and then the preset vehicle information list is searched according to the frame number of the video frame picture , It is determined whether the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, if it exists, step 103 is executed; if it does not exist, step 105 is executed. For example, if the frame number of the video frame picture to be detected is 123, then look up in the preset vehicle information table whether there is vehicle information of the video frame picture with the frame number 122. If it exists, then according to the video frame picture The vehicle information and the vehicle information of the previous video frame picture corresponding to the video frame picture are further judged whether there is a vehicle parking violation.

103. Calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture.

For the embodiment of the present application, in order to further determine whether the two cars in the video frame picture are illegal vehicles, the calculation of the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is specifically It includes: using a preset Euclidean distance algorithm to calculate the similarity between the feature vector of the vehicle in the video frame picture and the feature vector of the vehicle in the previous video frame picture, and the bounding box information of the vehicle in the video frame picture The similarity with the bounding box information of the vehicle in the previous video frame.

104. When the degree of similarity is less than a preset threshold, determine that the vehicle in the video frame picture violates a stop.

For the present embodiment, when the similarity is less than a preset threshold, determining that the vehicle in the video frame picture has violated parking includes: when the feature vector of the vehicle in the video frame picture is the same as the previous video The similarity of the feature vector of the vehicle in the frame picture is less than the first preset threshold, and the similarity between the bounding box information of the vehicle in the video frame picture and the bounding box information of the vehicle in the previous video frame picture is less than the first preset threshold. 2. When the threshold is preset, it is determined that the vehicle in the video frame picture has violated a stop. For example, the bounding box information of the vehicle in the video frame picture to be detected is M1 (x1, y1, w1, h1), and the vehicle feature vector is m1 (x ₁₁ , x ₁₂ ,..., x _1n ). The corresponding bounding box information of the vehicle in the previous frame of picture is N1(x2,y2,w2,h2), and the vehicle feature vector is n1(x ₂₁ , x ₂₂ ,..., x _2n ), if |x2-x1|<f1,|y2-y1|<f2,|w2-w1|<f3,|h2-h1|<f4 and the Euclidean distance between m1 and n1 is less than f5, it is determined that M1 and N1 belong to the same vehicle and are in a stopped state, That is, the vehicle has a parking violation, the Euclidean distance formula between M1 and N1 is as follows:

Among them, the preset thresholds f1, f2, f3, f4 and f5 can be determined statistically according to the real environment.

105. Determine whether the vehicle in the video frame picture is in the target lane area according to the vehicle information of the video frame picture.

Wherein, the target lane area is the lane line area where the vehicle is traveling on the road. For this embodiment of the present application, if the vehicle information of the previous video frame picture corresponding to the video frame picture does not exist in the preset vehicle information list, it will Set the vehicle information in the video frame picture extracted by the vehicle detection model to determine whether the vehicle in the video frame picture is in the target lane area, specifically, use the preset lane line detection algorithm to identify the lane line area in the video frame picture, According to the identified lane line area and the vehicle information in the extracted video frame picture, it is determined whether the vehicle in the video frame picture is within the lane line area.

106. Detect whether there is a vehicle in the next video frame picture corresponding to the video frame picture.

The embodiment of the application provides a vehicle parking violation detection method based on video frame picture analysis. Compared with the current method of monitoring vehicles through monitoring equipment installed on each road surface, the application can detect the video corresponding to the target no-parking area. Whether there is a vehicle in the frame picture; when a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is a previous video corresponding to the video frame picture in the preset vehicle information list The vehicle information of the frame picture, the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; if there is a previous one corresponding to the video frame picture in the preset vehicle information list The vehicle information of the video frame picture is calculated, and the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; at the same time, when the similarity is less than a preset threshold, it is determined that the The vehicle parking violation in the video frame picture can improve the recognition rate of parking violation vehicles in the lane, and at the same time, it can find the parking violation vehicles in time, assist the relevant departments to find the violation vehicles, improve the efficiency of finding the violation vehicles, thereby reducing Hidden traffic hazards caused by illegally parked vehicles.

Further, in order to better explain the above-mentioned process of detecting illegally parked vehicles in the video frame picture, as a refinement and extension of the above-mentioned embodiment, an embodiment of the present application provides another vehicle illegally detecting method, as shown in FIG. 2 As shown, the method includes:

201. Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area.

For this embodiment of the application, the preset vehicle detection model is used to perform vehicle detection on the video frame picture to be detected. If there is a vehicle in the video frame picture to be detected, step 202 is executed; if there is no vehicle in the video frame picture to be detected, Step 207 is executed, which is to continue to detect whether there is a vehicle in the next video frame picture. In addition, in order to detect whether there is a vehicle in the video frame picture, the preset vehicle detection model may be a first preset vehicle detection model, and the step 201 specifically includes: inputting the video frame picture corresponding to the target no-parking area into the first The vehicle detection model is preset for vehicle detection. Specifically, the first preset vehicle detection model may be a preset yolo V3 vehicle detection model. Although there is a yolo V3 model for vehicle detection in the prior art, in order to ensure the vehicle detection model The accuracy of the yolo V3 model is trained based on the video frame pictures obtained from the aerial video to obtain the preset yolo V3 vehicle detection model. Furthermore, the video frame pictures to be detected are input to the preset yolo V3 vehicle detection model for vehicle detection. Detect and determine whether there is a vehicle in the video frame picture, if it exists, execute step 202; if it does not exist, execute step 207.

At the same time, the preset vehicle detection model may also be a second preset vehicle detection model, and the step 201 specifically includes: inputting the video frame picture corresponding to the target no-parking area into the second preset vehicle detection model for vehicle detection. Detection, specifically, the second preset vehicle detection model may be a preset Mask R-CNN vehicle detection model. The preset Mask R-CNN vehicle detection model mainly includes three modules, namely, a full convolutional network, ROIAlign and Faster. -rcnn, the full convolutional network model of the full convolutional network has a total of 8 convolutional layers; the ROIAlign module will traverse each candidate area, keeping the floating-point number boundary without quantization, and then divide the candidate area into several units, each The boundary of the unit is not quantified, and four coordinate positions are fixed in each unit, and the values of these four positions are calculated by bilinear interpolation, and then the maximum pooling operation is performed; the Faster-rcnn module is mainly used To quickly generate candidate regions through the RPN network, the structure in front of the RPN network is the structure before the last layer of the ZF network, followed by the convolution layer with the convolution kernel of 3*3, and finally the convolution kernel is 1*1 The output of the convolutional layer is divided into two paths, one output is the probability of the target and non-target, and the other output is the four parameters of the target bounding box, which are the center coordinates, length and width of the bounding box. In addition, in order to improve the detection accuracy of the model, the existing Mask R-CNN model is trained according to the acquired aerial video frame pictures to obtain the preset Mask R-CNN vehicle detection model. Further, input the video frame pictures to be detected Perform vehicle detection to the preset Mask R-CNN vehicle detection model, and determine whether there is a vehicle in the video frame picture, if it exists, perform step 202; if it does not exist, perform step 207.

202. Extract the vehicle information of the vehicle from the video frame picture, and determine whether the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list.

Wherein, the preset vehicle information list stores the frame number of the video frame picture and its corresponding vehicle information. The vehicle information includes the bounding box information of the vehicle in the video frame picture and the feature vector of the vehicle. For the implementation of this application For example, in order to obtain the bounding box information of the vehicle and the feature vector of the vehicle in the video frame picture, when the preset vehicle detection model is the first preset vehicle detection model, step 202 specifically includes: Extracting the vehicle information of the vehicle includes: when the first preset vehicle detection model detects that there is a vehicle in the video frame picture, extracting a feature vector of the vehicle, and outputting the bounding box information of the vehicle; The feature vector of the vehicle and the bounding box information where the vehicle is located are determined as the vehicle information of the vehicle. For example, when the first preset vehicle detection model is the preset yolo V3 vehicle detection model, and the preset yolo V3 vehicle detection model detects that there is a car in the video frame picture, the output bounding box information of the vehicle is M( x, y, w, h) and the feature vector m of the vehicle, where x and y represent the coordinate information of the center point of the bounding box of the detected vehicle, w and h represent the size of the bounding box of the detected vehicle, and m is a 1024-dimensional feature vector.

In addition, when the preset vehicle detection model is the second preset vehicle detection model, step 202 specifically includes: said extracting the vehicle information of the vehicle from the video frame picture includes: when the second preset vehicle is detected When the model detects that there is a vehicle in the video frame picture, it extracts the feature vector of the vehicle, and outputs the bounding box information of the vehicle and the mask of the vehicle; the feature vector of the vehicle, the vehicle The bounding box information and the mask information of the vehicle are determined to be the vehicle information of the vehicle. For example, when the second preset vehicle detection model is the preset Mask R-CNN vehicle detection model, and the preset Mask R-CNN vehicle detection model detects that there is a car in the video frame picture, the bounding box of the vehicle is output The information is N(x,y,w,h), the feature vector n of the vehicle and the mask of the vehicle, where x and y represent the coordinate information of the center point of the bounding box of the detected vehicle, and w and h represent the bounding box of the detected vehicle The size of n is a 2048-dimensional feature vector.

Further, if the preset vehicle detection model detects that there is a vehicle in the video frame picture, the vehicle information of the vehicle in the video frame picture is extracted, and the frame number of the video frame picture is obtained, according to the frame number of the video frame picture Look up the preset vehicle information list, and determine whether there is vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, if it exists, execute step 203; if it does not exist, execute step 205.

203. Calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture by using a preset Euclidean distance algorithm.

For the embodiment of the present application, a preset Euclidean distance algorithm is used to calculate the similarity between the feature vector of the vehicle in the video frame picture and the feature vector of the vehicle in the previous video frame picture, and the location of the vehicle in the video frame picture. The similarity between the bounding box information and the bounding box information of the vehicle in the previous video frame picture. Specifically, the vehicle bounding box information of the video frame picture to be detected is M1 (x1, y1, w1, h1), and the vehicle The feature vector is m1 (x ₁₁ , x ₁₂ ,..., x _1n ), the vehicle bounding box information of the previous picture corresponding to the video frame picture is N1 (x2, y2, w2, h2), and the vehicle feature vector is n1 (x ₂₁ , x ₂₂ ,..., x _2n ), the similarity between the feature vector of the vehicle in the video frame picture and the feature vector of the vehicle in the previous video frame picture is

The similarity between the bounding box information of the vehicle in the video frame picture and the bounding box information of the vehicle in the previous video frame picture is |x2-x1|,|y2-y1|,|w2-w1|,|h2-h1| .

204. When the degree of similarity is less than a preset threshold, determine that the vehicle in the video frame picture violates a stop.

For example, the vehicle bounding box information of the video frame picture to be detected is M1 (x1, y1, w1, h1), and the vehicle feature vector is m1 (x ₁₁ , x ₁₂ ,..., x _1n ), which corresponds to the video frame picture The vehicle bounding box information of the previous picture is N1(x2,y2,w2,h2), and the vehicle feature vector is n1(x ₂₁ , x ₂₂ ,..., x _2n ), if |x2-x1|<f1,|y2 -y1|<f2,|w2-w1|<f3,|h2-h1|<f4 and the Euclidean distance between m1 and n1 is less than f5, then it is determined that M1 and N1 belong to the same vehicle and are in a stopped state, that is, the vehicle exists In case of parking violation, if any of the above conditions does not meet the requirements, it is determined that there is no vehicle parking violation in the target no-parking area. Further, after comparing M1 and N1, no vehicle parking violation is found temporarily, continue to compare the vehicle information M1 of the video frame picture with the vehicle information N2, N3, ... of the previous video frame picture, the comparison process is as above, Until the vehicle information M1 of the video frame picture is compared with the vehicle information N2, N3,... of the previous video frame picture, it is finally determined whether the vehicle with the vehicle information M1 in the video frame picture has a parking violation, and further If there is still vehicle information M2 in the video frame picture, continue to compare M2 with all the vehicle information N1, N2, N3, ... in the previous video frame picture, the comparison process is as above, and finally the video frame picture The vehicle information for M2 is whether there is a parking violation.

205. According to the vehicle information of the video frame picture, determine whether the vehicle in the video frame picture is in the target lane area.

For the embodiment of this application, according to the vehicle information extracted by the preset vehicle detection model, it is determined whether the vehicle in the video frame picture is in the target lane area, if yes, go to step 206; if not, go to step 207, and further In order to determine whether the vehicle is in the lane area, when the first preset vehicle detection model is the preset yolo V3 vehicle detection model, step 205 specifically includes: using a preset lane detection algorithm to perform lane detection on the video frame picture, Obtain the target lane area in the video frame picture; encode the target lane area in the video frame picture as 1, and encode the outside of the target area as 0, and put the vehicle in the video frame picture inside the bounding box The code is 1, the outer code of the bounding box where the vehicle is located is 0, and the coded regions with codes 0, 1, and 2 in the video frame picture are obtained; the number of coded regions with a code of 2 and the coded area with a code of 1 are counted The ratio of the number; if the ratio is greater than the preset ratio threshold, it is determined that the vehicle in the video frame picture is in the target lane area. At the same time, when the second preset vehicle detection model is the preset Mask R-CNN vehicle detection model, step 205 specifically includes: using a preset lane detection algorithm to perform lane detection on the video frame picture to obtain the video frame The target lane area in the picture; the target lane area in the video frame picture is internally coded as 1, the outside code of the target lane area is 0, and the mask of the vehicle in the video frame picture is internally coded as 1, so If the external coding of the mask is 0, the coding regions of the video frame pictures whose coding are 0, 1, and 2 are obtained; the ratio of the number of coding regions whose coding is 2 to the number of coding regions whose coding is 1 is calculated; if the ratio is If it is greater than the preset ratio threshold, it is determined that the vehicle in the video frame picture is in the target lane area.

Specifically, the detailed process of using lane detection to find the lane area is as follows: first, perform edge detection on the video frame picture, use Gaussian filter to smooth the video frame picture and eliminate noise; then calculate each pixel in the video frame picture Point gradient strength and direction, then apply non-maximum value suppression to eliminate spurious effects caused by edge detection; in addition, apply dual threshold detection to determine the true and potential edges; finally, edge detection is completed by suppressing isolated weak edges , The edge picture corresponding to the video frame picture is obtained; after the edge picture is obtained, the Hough transform is used to perform straight line detection on the edge picture. The details are as follows: Assume that the straight line equation passing through the edge picture pixel (x, y) is: =px+q, where p is the slope and q is the intercept. It can also be rewritten as q=-px+y. This formula is a straight line passing through the point (p, q) in the parameter space PQ. The polar coordinate equation represents a straight line, as follows: λ=xcosθ+ysinθ, and then a two-dimensional array of parameter space (λ, θ) is established according to the equation. This array is equivalent to an accumulator, searching for the target pixel sequentially, because the RGB value is white , So the target pixel is a white pixel. For each target pixel, find its corresponding position in the parameter space according to the formula λ=xcosθ+ysinθ, then add 1 to the corresponding position of the accumulator, and then find the maximum value in the parameter space accumulator , its position ^{([lambda],, [theta],),} and finally according to the parameter space position ^{([lambda],, [theta],)} and the formula λ = xcosθ + ysinθ find linear parameters of the edges of the image corresponding to the determined straight line based on the straight line parameters by This determines the lane line in the edge image; finally, use the graphics expansion operation on the lane line in the edge image to connect the discrete lane lines to obtain the final lane line. The area between the lane lines is the lane area. Further, after identifying the lane area, if the vehicle detection model is the preset yolo V3 vehicle detection model, according to the lane area and the vehicle bounding box detected from the video frame picture, determine whether the vehicle in the video frame picture is in In the lane area, specifically, the video frame pictures of the detected lane area and the detected vehicle bounding box are respectively coded 0-1. For this video frame picture, the inner lane area is coded as 1, and the outer lane area is coded as 0. At the same time, the area in the vehicle bounding box is coded as 1, and the area outside the vehicle bounding box is coded as 0. The code of the video frame picture is obtained as 0, 1, and 2 respectively, and the number of areas coded as 2 in the picture and the code The number of regions is 1, and finally the percentage of the number of regions encoded as 2 to the number of regions encoded as 1 is calculated. If the percentage of the number of encoding 2 and the number of encoding 1 is greater than the preset ratio threshold, it means that the vehicle in the video frame picture is detecting In the detected lane area, if the percentage is less than or equal to the preset ratio threshold, it means that the vehicle in the video frame picture is not in the detected lane area. Further, if the vehicle detection model is the preset Mask R-CNN model, the video frame picture of the detected lane area and the mask image of the vehicle are respectively coded 0-1. For the video frame picture of the detected lane area, the The inner code of the lane area is 1, and the outer code of the lane area is 0. For the mask of the vehicle, the area within the vehicle contour displayed in the vehicle mask is coded as 1, and the area outside the vehicle contour is coded as 0, and the specific video frame is determined The process of whether the vehicle in the picture is in the lane area is the same as the above process.

206. Obtain the frame number of the video frame picture, and correspondingly store the frame number and the vehicle information of the video frame picture in the preset vehicle information list.

For the embodiment of the present application, if it is determined that the vehicle in the video frame picture is within the detected lane area, then the vehicle information of the video frame picture and the frame number encoding of the video frame picture are stored in the preset vehicle information list; if It is determined that the vehicle in the video frame picture is not in the detected lane area, then another frame of the video frame picture is input again, and whether there is a vehicle in the picture is detected.

207. Detect whether there is a vehicle in the next video frame picture corresponding to the video frame picture.

The embodiment of the application provides another vehicle parking violation detection method based on the analysis of video frames. Compared with the current method of monitoring the vehicle through the monitoring equipment installed on each road surface, the application can detect the target no-parking area Whether there is a vehicle in the video frame picture; when a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is judged whether there is a previous one corresponding to the video frame picture in the preset vehicle information list The vehicle information of the video frame picture, the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; if the preset vehicle information list contains the previous image corresponding to the video frame picture For the vehicle information of a video frame picture, the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; at the same time, when the similarity is less than a preset threshold, it is determined The vehicle in the video frame picture is illegally parked, which can improve the recognition rate of illegally parked vehicles on the lane, and at the same time can detect illegally parked vehicles in time, assist relevant departments to find illegally parked vehicles, improve the efficiency of finding illegally parked vehicles, and reduce The hidden traffic hazards caused by illegally parked vehicles.

Further, as a specific implementation of FIG. 1, an embodiment of the present application provides a vehicle parking violation detection device based on video frame picture analysis. As shown in FIG. 3, the device includes: a detection unit 31, an extraction unit 32, and a calculation Unit 33 and determining unit 34.

The detection unit 31 may be used to detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area. The detection unit 31 is a main functional module of the device for detecting whether there is a vehicle in the video frame picture corresponding to the target no-parking area.

The extracting unit 32 may be used to extract vehicle information of the vehicle from the video frame picture when a vehicle is detected, and determine whether there is a front-end corresponding to the video frame picture in the preset vehicle information list. Vehicle information of a video frame picture. When a vehicle is detected in the device, the extraction unit 32 extracts the vehicle information of the vehicle from the video frame picture, and determines whether there is a front-end corresponding to the video frame picture in the preset vehicle information list. The main function module of the vehicle information of a video frame picture is also the core module.

The calculation unit 33 may be configured to, if the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, calculate the vehicle information of the previous video frame picture and all the vehicle information. The similarity of the vehicle information of the video frame pictures. The calculation unit 33 calculates the vehicle information of the previous video frame picture and the vehicle information of the previous video frame picture if the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list. The main functional module that describes the similarity of the vehicle information of the video frame picture is also the core module.

The determining unit 34 may be used to determine that the vehicle in the video frame picture has violated a stop when the similarity is less than a preset threshold. The determining unit 34 is a main functional module of the device for determining a vehicle in the video frame picture when the similarity is less than a preset threshold.

For the embodiment of the present application, if the vehicle information of the previous video frame picture corresponding to the video frame picture does not exist in the preset vehicle information list, the device further includes: a determination unit 35 and a storage unit 36, as shown in FIG. 4 Show.

The determining unit 35 may be configured to determine the video according to the vehicle information of the video frame picture if there is no vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list Whether the vehicle in the frame picture is in the target lane area.

The storage unit 36 may be used to obtain the frame number of the video frame picture if the vehicle in the video frame picture is in the target lane area, and combine the frame number with the vehicle information of the video frame picture Correspondingly stored in the preset vehicle information list.

Further, in order to calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture, the calculation unit 33 may be specifically used to calculate the vehicle information of the previous video frame picture by using a preset Euclidean distance algorithm The similarity between the information and the vehicle information of the video frame picture.

In a specific application scenario, the preset vehicle detection model is the first preset vehicle detection model, and the detection unit 31 may be specifically used to input the video frame picture corresponding to the target no-parking area to the first preset vehicle detection model. The model performs vehicle detection.

The extraction unit 32 includes: an extraction module and a determination module, and the extraction module may be used to extract a feature vector of the vehicle when the first preset vehicle detection model detects that there is a vehicle in the video frame picture , And output the bounding box information of the vehicle.

The determining module may be used to determine the feature vector of the vehicle and the bounding box information where the vehicle is located as the vehicle information of the vehicle.

In a specific application scenario, the preset vehicle detection model is a second preset vehicle detection model, and the detection unit 31 may also be specifically used to input the video frame picture corresponding to the target no-parking area to the second preset vehicle The detection model performs vehicle detection.

The extraction module may also be used to extract the feature vector of the vehicle when the second preset vehicle detection model detects that there is a vehicle in the video frame picture, and output the bounding box information of the vehicle and The mask of the vehicle.

The determining module may also be used to determine the feature vector of the vehicle, the bounding box information of the vehicle and the mask information of the vehicle as the vehicle information of the vehicle.

Further, in order to determine whether the vehicle in the video frame picture is in the target lane area, the determination unit 35 includes: a detection module 351, an encoding module 352, a statistics module 353, and a determination module 354.

The detection module 351 may be used to perform lane detection on the video frame picture by using a preset lane detection algorithm to obtain the target lane area in the video frame picture.

The encoding module 352 may be used to encode the target lane area in the video frame picture as 1 internally, the target area externally as 0, and encode the inner bounding box of the vehicle in the video frame picture as 1 The outer code of the bounding box where the vehicle is located is 0, and the coded regions of the video frame picture whose codes are respectively 0, 1, and 2 are obtained.

The statistics module 353 can be used to count the ratio of the number of coding regions coded as 2 to the number of coding regions coded as 1.

The determining module 354 may be configured to determine that the vehicle in the video frame picture is within the target lane area if the ratio is greater than a preset ratio threshold.

In addition, the detection module 351 can also be used to perform lane detection on the video frame picture by using a preset lane detection algorithm to obtain the target lane area in the video frame picture.

The encoding module 352 may also be used to internally encode the target lane area in the video frame picture to 1, and the external code to 0 in the target lane area, and to internally encode the mask of the vehicle in the video frame picture as 1. The outer code of the mask is 0, and the coded regions whose codes are 0, 1, and 2 in the video frame picture are obtained.

The statistics module 353 can also be used to count the ratio of the number of coding regions coded as 2 to the number of coding regions coded as 1.

The determining module 354 may also be used to determine that the vehicle in the video frame picture is within the target lane area if the ratio is greater than a preset ratio threshold.

It should be noted that, for other corresponding descriptions of the functional modules involved in the vehicle parking violation detection device based on video frame picture analysis provided by the embodiments of the present application, please refer to the corresponding description of the method shown in FIG. 1, which will not be repeated here. .

Based on the above method shown in FIG. 1, correspondingly, an embodiment of the present application also provides a computer non-volatile readable storage medium, on which computer readable instructions are stored, when the computer readable instructions are executed by the processor The following steps are implemented: detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area; when a vehicle is detected, extract the vehicle information of the vehicle from the video frame picture, and determine whether the preset vehicle information list is There is vehicle information of the previous video frame picture corresponding to the video frame picture, and the preset vehicle information list stores the frame number of the video frame picture and its corresponding vehicle information; if the preset vehicle information list If there is vehicle information of the previous video frame picture corresponding to the video frame picture, the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; when the similarity is less than When the threshold is preset, it is determined that the vehicle in the video frame picture has violated a stop.

Based on the above-mentioned method shown in FIG. 1 and the embodiment of the apparatus shown in FIG. 3, an embodiment of the present application also provides a physical structure diagram of a computer device. As shown in FIG. 5, the computer device includes: a processor 41, The memory 42 and the computer-readable instructions stored on the memory 42 and that can run on the processor, wherein the memory 42 and the processor 41 are both set on the bus 43, when the processor 41 executes the computer-readable instructions, the following is achieved Step: Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area; when a vehicle is detected, extract the vehicle information of the vehicle from the video frame picture, and determine whether the preset vehicle information list exists and The vehicle information of the previous video frame picture corresponding to the video frame picture, and the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; if there is one in the preset vehicle information list The vehicle information of the previous video frame picture corresponding to the video frame picture is calculated, and the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; when the similarity is less than a preset When the threshold is set, it is determined that the vehicle in the video frame picture has violated a stop.

Through the technical solution of the present application, it can be detected whether there is a vehicle in the video frame picture corresponding to the target no-parking area; when a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and the preset vehicle is determined Whether there is vehicle information of the previous video frame picture corresponding to the video frame picture in the information list, the preset vehicle information list stores the frame number of the video frame picture and its corresponding vehicle information; Assuming that the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the vehicle information list, the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; and At the same time, when the similarity is less than the preset threshold, it is determined that the vehicle in the video frame picture has violated parking, which can improve the recognition rate of parking vehicles on the lane, and at the same time, it can detect the parking vehicles in time and assist relevant departments in searching The illegally parked vehicles improves the efficiency of finding the illegally parked vehicles, thereby reducing the hidden traffic hazards caused by the illegally parked vehicles.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices. Above, alternatively, they can be implemented with computer-readable instruction codes executable by a computing device, so that they can be stored in a storage device for execution by the computing device, and in some cases, can be different from this The steps shown or described are executed in the order in which they are shown, or they are respectively fabricated into individual integrated circuit modules, or multiple modules or steps of them are fabricated into a single integrated circuit module for implementation. In this way, this application is not limited to any specific hardware and software combination.

The above descriptions are only preferred embodiments of the application, and are not used to limit the application. For those skilled in the art, the application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims (20)

1. A vehicle parking violation detection method based on video frame picture analysis, which is characterized in that it includes:

   Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

   When a vehicle is detected, the vehicle information of the vehicle is extracted from the video frame picture, and it is determined whether there is the vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, so The preset vehicle information list stores the frame number of the video frame picture and the corresponding vehicle information;

   If the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list, the vehicle information of the previous video frame picture is calculated to be similar to the vehicle information of the video frame picture degree;

   When the similarity is less than a preset threshold, it is determined that the vehicle in the video frame picture has violated a stop.
2. The method according to claim 1, wherein after said determining whether there is vehicle information of the previous video frame picture corresponding to the video frame picture in the preset vehicle information list, the method further comprises:

   If the previous video frame picture corresponding to the video frame picture does not exist in the preset vehicle information list, determine whether the vehicle in the video frame picture is in the target lane area according to the vehicle information of the video frame picture Inside;

   If the vehicle in the video frame picture is within the target lane area, the frame number of the video frame picture is obtained, and the frame number and the vehicle information of the video frame picture are correspondingly stored in the preset Vehicle information list.
3. The method according to claim 1, wherein the calculating the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture comprises:

   A preset Euclidean distance algorithm is used to calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture.
4. The method according to claim 1, wherein the detecting whether there is a vehicle in the video frame picture corresponding to the target no-parking area comprises:

   Input the video frame picture corresponding to the target no parking area into the first preset vehicle detection model for vehicle detection;

   The extracting vehicle information of the vehicle from the video frame picture includes:

   When the first preset vehicle detection model detects that there is a vehicle in the video frame picture, extract the feature vector of the vehicle, and output the bounding box information of the vehicle;

   The feature vector of the vehicle and the bounding box information where the vehicle is located are determined as the vehicle information of the vehicle.
5. The method according to claim 1, wherein the detecting whether there is a vehicle in the video frame picture corresponding to the target no-parking area comprises:

   Input the video frame picture corresponding to the target no parking area into the second preset vehicle detection model for vehicle detection;

   The extracting vehicle information of the vehicle from the video frame picture includes:

   When the second preset vehicle detection model detects that there is a vehicle in the video frame picture, extract the feature vector of the vehicle, and output the bounding box information of the vehicle and the mask of the vehicle;

   The feature vector of the vehicle, the bounding box information of the vehicle and the mask information of the vehicle are determined as the vehicle information of the vehicle.
6. The method according to claim 4, wherein the determining whether the vehicle in the video frame picture is in the target lane area according to the vehicle information of the video frame picture comprises:

   Performing lane detection on the video frame picture by using a preset lane detection algorithm to obtain the target lane area in the video frame picture;

   Encode the target lane area inside the video frame picture as 1, and encode the outside of the target area as 0, and encode the inside bounding box of the vehicle in the video frame picture as 1, and the vehicle outside the bounding box Encoding is 0, and the encoding regions of the video frame pictures with encodings of 0, 1, and 2 are obtained;

   Count the ratio of the number of coding regions with a code of 2 to the number of coding regions with a code of 1;

   If the ratio is greater than a preset ratio threshold, it is determined that the vehicle in the video frame picture is in the target lane area.
7. The method according to claim 5, wherein the determining whether the vehicle in the video frame picture is in the target lane area according to the vehicle information of the video frame picture comprises:

   Performing lane detection on the video frame picture by using a preset lane detection algorithm to obtain the target lane area in the video frame picture;

   In the video frame picture, the target lane area is internally coded as 1, and the target lane area is externally coded as 0, and the mask of the vehicle in the video frame picture is internally coded as 1, and the mask outside is coded as 0 , Obtain the coding regions of the video frame pictures whose codes are respectively 0, 1, and 2;

   Count the ratio of the number of coding regions with a code of 2 to the number of coding regions with a code of 1;

   If the ratio is greater than a preset ratio threshold, it is determined that the vehicle in the video frame picture is in the target lane area.
8. A vehicle parking violation detection device based on video frame picture analysis, characterized in that it comprises:

   The detection unit is used to detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area;

   The extraction unit is configured to extract vehicle information of the vehicle from the video frame picture when a vehicle is detected, and determine whether there is a previous video frame picture corresponding to the video frame picture in the preset vehicle information list The vehicle information of the video frame picture and the corresponding vehicle information are stored in the preset vehicle information list;

   The calculation unit is configured to calculate the vehicle information of the previous video frame picture and the video frame picture if the vehicle information of the previous video frame picture corresponding to the video frame picture exists in the preset vehicle information list The similarity of the vehicle information;

   The determining unit is configured to determine that the vehicle in the video frame picture illegally stops when the similarity is less than a preset threshold.
9. The device according to claim 8, wherein the device further comprises:

   The determining unit is configured to determine the vehicle in the video frame picture according to the vehicle information of the video frame picture if there is no previous video frame picture corresponding to the video frame picture in the preset vehicle information list Whether it is in the target lane area;

   A storage unit, configured to obtain the frame number of the video frame picture if the vehicle in the video frame picture is in the target lane area, and store the frame number and the vehicle information of the video frame picture correspondingly To the preset vehicle information list.
10. 8. The device according to claim 8, wherein the calculation unit is specifically configured to use a preset Euclidean distance algorithm to calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture .
11. 8. The device according to claim 8, wherein the detection unit is specifically configured to input the video frame picture corresponding to the target no-parking area into the first preset vehicle detection model for vehicle detection; the extraction unit includes : An extraction module, used to extract the feature vector of the vehicle when the first preset vehicle detection model detects a vehicle in the video frame picture, and output the bounding box information where the vehicle is located; a determination module, It is used to determine the feature vector of the vehicle and the bounding box information where the vehicle is located as the vehicle information of the vehicle.
12. 8. The device according to claim 8, wherein the detection unit is specifically further configured to input the video frame picture corresponding to the target no-parking area into the second preset vehicle detection model for vehicle detection; the extraction module, It is also used to extract the feature vector of the vehicle when the second preset vehicle detection model detects that there is a vehicle in the video frame picture, and output the bounding box information of the vehicle and the mask of the vehicle The determining module is also used to determine the feature vector of the vehicle, the bounding box information of the vehicle and the mask information of the vehicle as the vehicle information of the vehicle.
13. The method according to claim 11, wherein the determining unit comprises:

    The detection module is used to perform lane detection on the video frame picture using a preset lane detection algorithm to obtain the target lane area in the video frame picture; the encoding module is used to internally encode the target lane area in the video frame picture Is 1, the outer code of the target area is 0, the inner code of the bounding box of the vehicle in the video frame picture is 1, and the outer code of the bounding box of the vehicle is 0, and the code in the video frame picture is obtained The coding regions are respectively 0, 1, and 2; the statistics module is used to count the ratio of the number of coding regions coded as 2 to the number of coding regions coded as 1; the determining module is used to if the ratio is greater than a preset ratio threshold, It is determined that the vehicle in the video frame picture is in the target lane area.
14. The method according to claim 12, wherein:

    The detection module is further configured to use a preset lane detection algorithm to perform lane detection on the video frame picture to obtain the target lane area in the video frame picture; the encoding module is also used to convert the video frame picture The inner code of the middle target lane area is 1, the outer code of the target lane area is 0, and the mask of the vehicle in the video frame picture is coded as 1, and the mask outer code is 0, to obtain the video frame The coded areas in the picture are respectively 0, 1, and 2; the statistical module is also used to count the ratio of the number of coded areas with code 2 to the number of coded areas with code 1; the determining module is also used for If the ratio is greater than a preset ratio threshold, it is determined that the vehicle in the video frame picture is in the target lane area.
15. A computer nonvolatile readable storage medium having computer readable instructions stored thereon, wherein the computer readable instructions are executed by a processor to realize a vehicle parking violation detection method based on video frame picture analysis, including :

    Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area; when a vehicle is detected, extract the vehicle information of the vehicle from the video frame picture, and determine whether there is a vehicle in the preset vehicle information list. The vehicle information of the previous video frame picture corresponding to the video frame picture, the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; The vehicle information of the previous video frame picture corresponding to the video frame picture, then the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; when the similarity is less than a preset threshold , It is determined that the vehicle in the video frame picture has illegally stopped.
16. The computer non-volatile readable storage medium according to claim 15, wherein when the computer readable instruction is executed by a processor, the calculation of the vehicle information of the previous video frame picture and the video The similarity of the vehicle information of the frame picture includes:

    A preset Euclidean distance algorithm is used to calculate the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture.
17. The computer non-volatile readable storage medium according to claim 15, wherein the computer readable instruction is executed by the processor to realize whether there is a vehicle in the video frame picture corresponding to the detection target no parking area, include:

    Inputting the video frame picture corresponding to the target no-parking area into the first preset vehicle detection model for vehicle detection; said extracting the vehicle information of the vehicle from the video frame picture includes: when the first preset vehicle is detected When the model detects that there is a vehicle in the video frame picture, it extracts the feature vector of the vehicle, and outputs the bounding box information of the vehicle; the feature vector of the vehicle and the bounding box information of the vehicle are determined Is the vehicle information of the vehicle.
18. A computer device, including a memory, a processor, and computer-readable instructions stored on the memory and running on the processor, wherein the computer-readable instructions are executed by the processor to realize image analysis based on video frames Vehicle parking violation detection methods, including:

    Detect whether there is a vehicle in the video frame picture corresponding to the target no-parking area; when a vehicle is detected, extract the vehicle information of the vehicle from the video frame picture, and determine whether there is a vehicle in the preset vehicle information list. The vehicle information of the previous video frame picture corresponding to the video frame picture, the frame number of the video frame picture and its corresponding vehicle information are stored in the preset vehicle information list; The vehicle information of the previous video frame picture corresponding to the video frame picture, then the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture is calculated; when the similarity is less than a preset threshold , It is determined that the vehicle in the video frame picture has illegally stopped.
19. The computer device according to claim 18, wherein when the computer-readable instructions are executed by a processor, the calculation of the vehicle information of the previous video frame picture is similar to the vehicle information of the video frame picture The degree includes: calculating the similarity between the vehicle information of the previous video frame picture and the vehicle information of the video frame picture by using a preset Euclidean distance algorithm.
20. The computer device according to claim 18, wherein when the computer-readable instruction is executed by the processor, the detection of whether there is a vehicle in the video frame picture corresponding to the target no-parking area comprises: setting the target no-parking area The corresponding video frame picture is input to the first preset vehicle detection model for vehicle detection; said extracting the vehicle information of the vehicle from the video frame picture includes: when the first preset vehicle detection model detects the When there is a vehicle in the video frame picture, extract the feature vector of the vehicle and output the bounding box information of the vehicle; determine the feature vector of the vehicle and the bounding box information of the vehicle as the vehicle's Vehicle Information.

Images