A community vehicle management method and system based on AI vehicle recognition

Through multi-camera AI vehicle identification technology, the evaluation speed and flow rate of the community vehicle are calculated and the congestion index is predicted, which solves the problem of misjudgment of vehicle congestion monitoring in community vehicle management, and improves management accuracy and efficiency.

CN118587908BActive Publication Date: 2024-11-08ROPEOK TECHNOLOGY GROUP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411081394.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-08
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

The existing traffic flow prediction model ignores the characteristics of slow vehicle travel speed and immediate parking in the prediction of community vehicle congestion, resulting in misjudgment and increased management costs, making it difficult to accurately monitor the degree of vehicle congestion in the community.

Method used

Using AI-based vehicle identification method, video sets are obtained through multiple cameras, vehicle marking classification, vehicle flow calculation and dynamic density calculation, combined with preset camera layout information, calculate vehicle evaluation speed and vehicle flow rate, build vehicle flow feature vectors, and predict Vehicle congestion index, used to control community gates.

Benefits of technology

The utilization rate of vehicle speed information and the sensitivity of community vehicle information are improved, the degree of vehicle congestion is accurately reflected, and the vehicle entry into the community is reasonably controlled, so as to reduce vehicle congestion in the community.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118587908B_ABST
    Figure CN118587908B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of road control technology, and specifically to a community vehicle management method and system based on AI vehicle identification, including obtaining a video set of preset duration and marking it to obtain a labeled video set; calculating the community traffic rate, community vehicle dynamic density and community traffic flow corresponding to each camera area in each observation time period; and predicting based on this to obtain the predicted traffic feature vector corresponding to each camera area; calculating the community vehicle congestion index based on the predicted traffic feature vector corresponding to each camera area, and controlling the gate of the community according to the community vehicle congestion index. In the present invention, the calculation method of vehicle evaluation speed, community traffic flow and community vehicle dynamic density is optimized. The vehicle evaluation speed, community traffic flow and community vehicle dynamic density are predicted, and different gate control strategies are adopted according to the prediction results to reduce the occurrence of vehicle congestion in the community.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of road control, and in particular to a community vehicle management method and system based on AI vehicle recognition. Background Art

[0002] In recent years, with the rapid increase in the number of cars in cities, not only the management and construction of urban public roads needs to be strengthened, but also the management of vehicles entering and leaving the community is particularly important. Since the roads in the community are narrow, it is difficult to guide vehicles when they are congested. At the same time, there are many pedestrians in the community, and frequent vehicle congestion will greatly increase the possibility of safety accidents. Therefore, vehicle entry and exit management has become an important task in the daily affairs of residential communities with a large number of vehicles.

[0003] However, the current traffic flow prediction model is mainly used for the prediction of vehicle congestion on public roads, and usually only considers basic parameters such as traffic volume, driving speed, and traffic density. When predicting vehicle congestion in residential areas, the application often ignores the characteristics of slow vehicle speed and immediate parking in the residential area, which leads to misjudgment of the degree of vehicle congestion in the residential area and sends wrong signals to the gate at the entrance of the residential area, causing vehicles to be unable to enter the residential area in time, increasing the cost of manual management, and causing inconvenience to the owners and property management personnel of the residential area. Therefore, how to improve the accuracy of monitoring the degree of vehicle congestion in the residential area, reasonably control the vehicles entering the residential area, and reduce the occurrence of vehicle congestion in the residential area is an urgent problem to be solved in the current residential vehicle management. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a community vehicle management method and system based on AI vehicle identification to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In the first aspect, the present application provides a community vehicle management method based on AI vehicle identification, the method comprising: obtaining a video set of preset duration, the video set being obtained by shooting roads in different areas by at least three cameras, the preset duration being composed of several observation time periods, and the end time of the preset duration being the current time; marking and classifying the vehicles in the video set to obtain a labeled video set; calculating the traffic flow according to the preset camera acquisition frame rate and the labeled video set to obtain the community traffic flow corresponding to each camera area in each observation time period and the vehicle evaluation speed of each vehicle; performing dynamic density calculation according to the preset camera layout information in the community and the labeled video set to obtain The dynamic density of vehicles in the community corresponding to each camera area in each observation time period; the community traffic rate corresponding to each camera area in each observation time period is calculated based on the evaluated speeds of all the vehicles corresponding to each camera area in each observation time period, and the community traffic rate, the community vehicle dynamic density and the community traffic flow are recorded as the traffic feature vector in the observation time period; prediction is performed based on the traffic feature vector corresponding to each observation time period and a preset prediction model to obtain the predicted traffic feature vector corresponding to each camera area; the community vehicle congestion index is calculated based on the predicted traffic feature vector corresponding to each camera area, and the community gate is controlled according to the community vehicle congestion index.

[0006] In combination with the first aspect, in a possible implementation method, the vehicles in the video set are labeled and classified to obtain a labeled video set, including: identifying the vehicle license plate in each frame in the video set according to a preset YOLOv4 mathematical model to obtain the license plate information corresponding to each vehicle; sending the license plate information corresponding to each vehicle to a community vehicle management system to obtain feedback information; and classifying and marking the vehicles in each frame according to the feedback information to obtain a labeled video set.

[0007] In combination with the first aspect, in a possible implementation method, according to the preset camera acquisition frame rate and the traffic flow calculation on the label, the community traffic flow corresponding to each camera area in each observation time period and the vehicle evaluation speed of each vehicle are obtained, including: counting the total number of frames in which each vehicle appears in the camera area during the observation time period to obtain the vehicle frame number; calculating according to the length of the observation time period, the preset camera acquisition frame rate and the vehicle frame number to obtain the vehicle evaluation speed of each vehicle in each camera area; and counting the total number of vehicles entering and leaving each camera area during the observation time period to obtain the community traffic flow corresponding to each camera area.

[0008] In combination with the first aspect, in a possible implementation method, the vehicles on each frame of the road image in the labeled video set are classified as owner vehicles or non-owner vehicles, and dynamic density calculation is performed based on the preset camera layout information in the community and the labeled video set to obtain the community vehicle dynamic density corresponding to each camera area in each observation time period, including: querying the appearance status of real-time owner vehicles in the historical labeled video set respectively, the real-time owner vehicles are owner vehicles that appear in the labeled video set corresponding to the real-time observation time period, the end time of the real-time observation time period is the current time, the historical labeled video set is the labeled video set corresponding to multiple historical observation time periods, and the The historical observation time periods are at least two observation time periods within a preset time length except the real-time observation time period; the relative vehicle detention distance corresponding to each of the owner's vehicles in each frame of road image in the real-time observation time period is calculated according to the preset camera layout information in the community, the appearance status of each owner's vehicle and the label video set; the dynamic density of community vehicles corresponding to each camera area in the real-time observation time period is obtained according to the relative vehicle detention distance corresponding to each of the owner's vehicles in each frame of road image in the real-time observation time period and the relative vehicle detention distance corresponding to all the non-owner vehicles, and the relative vehicle detention distance corresponding to all the non-owner vehicles is a preset value.

[0009] In combination with the first aspect, in a possible implementation method, the relative vehicle detention distance corresponding to each owner vehicle in each frame of the road image in the real-time observation time period is obtained according to the preset camera layout information in the community, the appearance status of each owner vehicle and the label video set, including: if the appearance status of the owner vehicle is appeared, then the distance between the first camera and the second camera is extracted from the preset camera layout information in the community, and recorded as the vehicle entry distance, the first camera is the camera that shoots the owner vehicle, and the second camera is the camera located in the community parking lot; if the appearance status of the owner vehicle is not appeared, then the distance between the first camera and the third camera is extracted from the preset camera layout information in the community, and recorded as the vehicle exit distance, the third camera is the camera located at the community exit gate; the vehicle exit distance or the vehicle entry distance is input into the detention relative distance calculation formula to obtain the vehicle detention relative distance.

[0010] In combination with the first aspect, in a possible implementation method, the dynamic density of vehicles in the community corresponding to each camera area in the real-time observation time period is obtained according to the relative distance of vehicle detention corresponding to each of the owner vehicles and the relative distance of vehicle detention corresponding to all the non-owner vehicles in each frame of road image in the real-time observation time period, including: normalizing the relative distance of vehicle detention corresponding to all owner vehicles, and updating the relative distance of vehicle detention corresponding to all owner vehicles to the normalized result; accumulating the relative distance of vehicle detention corresponding to all owner vehicles and all non-owner vehicles corresponding to each frame of road image in the camera area, obtaining the vehicle density of the community corresponding to each frame of road image; calculating the relative distance of vehicle detention corresponding to each frame of road image according to the relative distance of vehicle detention corresponding to each owner vehicle; and calculating the relative distance of vehicle detention corresponding to each frame of road image according to the relative distance of vehicle detention corresponding to each non-owner vehicle. The vehicle density of each camera area is summarized to obtain the change of the cell vehicle density in each camera area within the observation time period; the mean value of the cell vehicle density corresponding to each frame of road image is calculated to obtain the mean value of the cell vehicle density in each camera area within the observation time period; a straight line fitting is performed on the change of the cell vehicle density in each camera area within the observation time period according to the time series relationship to obtain the slope of the fitting line corresponding to each camera area; the slopes of the fitting lines corresponding to all the camera areas are normalized to obtain the density coefficient corresponding to each camera area; the mean value and density coefficient of the cell vehicle density corresponding to each camera area within the observation time period are calculated to obtain the dynamic density of the cell vehicles corresponding to each camera area.

[0011] In combination with the first aspect, in a possible implementation, the community traffic flow rate in each camera area is obtained according to the calculation of all the vehicle evaluation speeds corresponding to each camera area, including: summarizing all the vehicle evaluation speeds corresponding to each camera area to obtain the vehicle evaluation speed change of each vehicle in the observation time period; calculating the mean according to the vehicle evaluation speed change of each vehicle in the observation time period to obtain the mean vehicle evaluation speed corresponding to each vehicle; performing straight line fitting on the vehicle evaluation speed change of each vehicle in the observation time period according to the time series relationship to obtain the fitting straight line slope corresponding to each vehicle; normalizing the fitting straight line slope corresponding to each vehicle to obtain the speed coefficient corresponding to each vehicle; calculating according to the speed coefficient corresponding to each vehicle and the mean vehicle evaluation speed to obtain the vehicle dynamic rate of each vehicle; calculating according to the appearance relationship of the vehicle in each camera area and the vehicle dynamic rate of each vehicle, respectively obtaining the vehicle dynamic rate mean and the vehicle dynamic rate standard deviation corresponding to each camera area; calculating according to the vehicle dynamic rate mean and the vehicle dynamic rate standard deviation corresponding to each camera area to obtain the community traffic flow rate in each camera area.

[0012] In combination with the first aspect, in a possible implementation method, prediction is performed based on the traffic feature vector corresponding to each observation time period and a preset prediction model to obtain a predicted traffic feature vector corresponding to each camera area, including: constructing a hypergraph for each observation time period respectively to obtain a community vehicle hypergraph corresponding to each observation time period, each node of the community vehicle hypergraph includes a traffic feature vector, and the spacing between the nodes of the community vehicle hypergraph and all corresponding camera areas of the nodes connected thereto is less than a preset distance; calculating the association matrix corresponding to each observation time period based on all the community vehicle hypergraphs; and inputting a preset number of community vehicle hypergraphs and the corresponding association matrices into a preset prediction model for prediction to obtain a predicted traffic feature vector corresponding to each camera area.

[0013] In combination with the first aspect, in a possible implementation method, a community vehicle congestion index is calculated based on the predicted traffic flow feature vector of the community, and the community gate is controlled according to the community vehicle congestion index, including: calculating the congestion index based on the congestion index calculation formula and the predicted traffic flow feature vector corresponding to each camera area to obtain the vehicle congestion index corresponding to each camera area; accumulating the vehicle congestion index corresponding to all the camera areas to obtain the community vehicle congestion index; and judging whether to allow vehicles to enter the community within each preset time period according to the community vehicle congestion index.

[0014] On the second aspect, the present application also provides a community vehicle management system based on AI vehicle identification, including: a camera module, used to obtain a video set of preset duration, the video set is obtained by shooting roads in different areas by at least three cameras, the preset duration is composed of several observation time periods, and the end time of the preset duration is the current time; a classification module, used to mark and classify the vehicles in the video set to obtain a labeled video set; a first calculation module, used to calculate the traffic flow according to the preset camera acquisition frame rate and the labeled video set, and obtain the community traffic flow corresponding to each camera area in each observation time period and the vehicle evaluation speed of each vehicle; a second calculation module, used to perform dynamic density calculation based on the preset camera layout information in the community and the labeled video set, to obtain to the dynamic density of vehicles in the community corresponding to each camera area in each observation time period; a third calculation module, used to calculate according to the evaluated speeds of all the vehicles corresponding to each camera area in each observation time period, to obtain the community traffic rate corresponding to each camera area in each observation time period, and record the community traffic rate, the community vehicle dynamic density and the community traffic flow as the traffic feature vector in the observation time period; a prediction module, used to predict according to the traffic feature vector corresponding to each observation time period and a preset prediction model, to obtain the predicted traffic feature vector corresponding to each camera area; a gate control module, used to calculate the community vehicle congestion index based on the predicted traffic feature vector corresponding to each camera area, and control the gate of the community according to the community vehicle congestion index.

[0015] The present invention has the following beneficial effects:

[0016] The present invention processes the collected video set and combines the camera acquisition frame rate and the camera layout information in the community to obtain the vehicle evaluation speed and the community traffic flow, wherein the vehicle evaluation speed and the community traffic flow calculated by combining the camera acquisition frame rate and the camera layout information in the community improve the utilization rate of the vehicle speed information and more accurately reflect the relative congestion degree of the vehicles in the camera area. Then, the dynamic density of the community vehicles that is more in line with the traffic characteristics of the community is calculated by using all the vehicle evaluation speeds corresponding to each camera area in each observation time period, thereby improving the utilization rate of the community vehicle information and the sensitivity of reflecting the congestion degree of the community. Finally, the vehicle evaluation speed, the community traffic flow and the dynamic density of the community vehicles are predicted, and different gate control strategies are adopted according to the prediction results to complete the control of the release of vehicles entering the community, improve the accuracy of monitoring the congestion degree of vehicles in the community, reasonably control the vehicles entering the community, and reduce the occurrence of vehicle congestion in the community. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a community vehicle management method based on AI vehicle identification provided by Embodiment 1 of the present invention;

[0019] Figure 2 A schematic diagram of the flow of step S2 provided in Example 1 of the present invention;

[0020] Figure 3 A schematic flow chart of step S3 provided in Example 1 of the present invention;

[0021] Figure 4 A schematic flow chart of step S4 provided in Example 1 of the present invention;

[0022] Figure 5 A schematic flow chart of step S5 provided in Example 1 of the present invention;

[0023] Figure 6 A schematic flow chart of step S6 provided in Example 1 of the present invention;

[0024] Figure 7 A schematic diagram of the flow of step S7 provided in Example 1 of the present invention;

[0025] Figure 8 This is a schematic diagram of the structure of the community vehicle management system based on AI vehicle identification described in Example 2 of the present invention. DETAILED DESCRIPTION

[0026] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a community vehicle management method and system based on AI vehicle recognition proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0027] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0028] Embodiment 1:

[0029] The following is a detailed description of a specific solution of a community vehicle management method based on AI vehicle identification provided by the present invention in conjunction with the accompanying drawings.

[0030] See also Figure 1 , which shows a flowchart of a community vehicle management method based on AI vehicle identification provided by an embodiment of the present invention, wherein the community vehicle management method based on AI vehicle identification provided by this embodiment includes step S1, step S2, step S3, step S4, step S5, step S6 and step S7.

[0031] S1. Obtain a video set of preset duration, where the video set is obtained by shooting roads in different areas by at least three cameras. The preset duration consists of several observation time periods, and the end time of the preset duration is the current time.

[0032] Specifically, in this embodiment, the camera can be a CCD camera or other commercial cameras, and no specific restrictions are made in this embodiment. In addition, the camera acquisition frequency is 50, each observation time period is 1 minute, and the preset duration is 30 minutes. In short, in this embodiment, a total of multiple 30-minute video segments are obtained. At the same time, the number of cameras in this embodiment is N, and the specific number can be determined according to the size of the cell to which it is applied. For example, in this embodiment, N is 100.

[0033] S2. Label and classify the vehicles in the video set to obtain a labeled video set.

[0034] Since the video captured by the camera contains a large amount of information, but the useful information in this embodiment is vehicle-related information, this step also includes steps S21, S22 and S23. Figure 2 , which shows a schematic flow chart of step S2 provided in an embodiment 1 of the present invention, including:

[0035] S21. Identify the vehicle license plate in each frame of the video set according to a preset YOLOv4 mathematical model to obtain the license plate information corresponding to each vehicle.

[0036] It should be noted that the method of using the YOLOv4 mathematical model is a prior art and will not be repeated in this embodiment.

[0037] S22: Send the license plate information corresponding to each vehicle to the community vehicle management system to obtain feedback information.

[0038] Among them, the community vehicle management system can compare the registered owner's license plate information with the hash search algorithm to obtain feedback results. At the same time, the feedback information can be that the license plate does not belong to the owner or the license plate belongs to the owner, or it can be no query or query results, etc. to distinguish whether the feedback license plate information belongs to a registered license plate in the community vehicle management system. No specific restrictions are made in this embodiment.

[0039] S23. Classify and label the vehicles in each frame according to the feedback information to obtain a labeled video set.

[0040] That is, in this step, each vehicle in each frame is marked as the owner's vehicle or the non-owner's vehicle by feedback information. At the same time, the license plate information of each vehicle can be marked for subsequent calculation by computer personnel in this field.

[0041] S3. Carry out traffic flow calculation according to the preset camera acquisition frame rate and label video set to obtain the community traffic flow corresponding to each camera area and the vehicle evaluation speed of each vehicle in each observation time period.

[0042] Specifically, the calculation step of the traffic flow of the community and the vehicle evaluation speed of each vehicle in each observation time period in this step includes step S31, step S32 and step S33. Figure 3 , which shows a schematic flow chart of step S3 provided in an embodiment 1 of the present invention, including:

[0043] S31. Count the total number of frames in which each vehicle appears in the camera area during the observation period to obtain the vehicle frame number.

[0044] S32. Obtain the vehicle evaluation speed of each vehicle in each camera area according to the duration of the observation time period, the preset camera acquisition frame rate and the vehicle frame number.

[0045] Specifically, in this step, the total number of frames is first calculated according to the length of the observation time period and the preset camera acquisition frame rate, and then the total number of frames is compared with the vehicle frame number to obtain the vehicle evaluation speed of each vehicle. Since the road lengths in the camera area are not much different, the more frames a community vehicle appears in the same camera area, the slower the driving speed of the community vehicle, and the smaller the calculated vehicle evaluation speed. And it should be noted that this step is a demonstrative calculation of the vehicle evaluation speed of each vehicle in each camera area in an observation time period. For each vehicle in each camera area in other observation time periods, the method of this step can also be used for calculation. At the same time, for the convenience of calculation, for some vehicles that only appear in one frame in the camera area in an observation time period, the vehicle evaluation speed of such vehicles is set to the vehicle default speed v, and the value of v in this solution is 1.

[0046] S33. Count the total number of vehicles entering and leaving each camera area during the observation period to obtain the community traffic flow corresponding to each camera area.

[0047] S4. Perform dynamic density calculation based on preset camera layout information in the community and the label video set to obtain the community vehicle dynamic density corresponding to each camera area in each observation time period.

[0048] In this embodiment, it is taken into account that the roads in the community are different from public roads, and there are generally multiple parking lots inside, and vehicles in the community can be parked in the parking lots immediately. When there are multiple owner vehicles on the community road that want to enter the parking lot or leave the community, the vehicle congestion level of the next community vehicle observation interval will be greatly reduced. However, the prior art ignores this feature of community vehicles in the process of calculating vehicle density, and its sensitivity in reflecting the degree of vehicle congestion in the community is insufficient, resulting in the failure to release vehicles that need to enter the community in time. In addition, most of the vehicles in the community are owner vehicles, and their parking space information is registered in the community vehicle management system. Using the vehicle information registered by the owner in the community vehicle management system and the location information of the parking lot, the probability and time of the owner's vehicle parking in and out of the community are analyzed to improve the subsequent prediction of the degree of vehicle congestion in the community. Therefore, in this embodiment, the dynamic density of community vehicles in the community is calculated to better reflect the instant parking characteristics of the community. Specifically, this step includes step S41, step S42 and step S43. Please refer to Figure 4 , which shows a schematic flow chart of step S4 provided in an embodiment 1 of the present invention, including:

[0049] S41. Query the appearance status of the real-time owner vehicle in the historical label video set respectively, wherein the real-time owner vehicle is the owner vehicle that appears in the label video set corresponding to the real-time observation time period, the end time of the real-time observation time period is the current time, the historical label video set is the label video set corresponding to multiple historical observation time periods, and the historical observation time period is at least two observation time periods within a preset duration except the real-time observation time period.

[0050] Specifically, take one of the owner's vehicles A as an example. In this step, you can choose to query whether the owner's vehicle appears at the gate at the entrance of the community within L observation time periods. If it appears, the appearance status of the owner's vehicle A is already appeared; if it does not appear, the appearance status of the owner's vehicle A is not appeared, where L can be 30.

[0051] S42, according to the preset camera layout information in the community, the appearance status of each owner's vehicle and the label video set, the relative vehicle detention distance corresponding to each owner's vehicle in each frame of road image in the real-time observation time period is obtained.

[0052] Among them, step S42 can be performed using the following steps.

[0053] S421. If the appearance status of the owner's vehicle is already appeared, the distance between the first camera and the second camera is extracted from the preset camera layout information in the community and recorded as the vehicle entry distance ens. The first camera is a camera for photographing the owner's vehicle, and the second camera is a camera located in the community parking lot.

[0054] It should be noted that the distance between the first camera and the second camera refers to the distance between the camera areas of the first camera and the second camera.

[0055] S422: If the owner's vehicle is not present, extract the distance between the first camera and the third camera from the preset camera layout information in the community and record it as the vehicle exit distance. , the third camera is a camera located at the gate at the exit of the community.

[0056] It should be noted that the distance between the first camera and the third camera refers to the distance between the camera areas of the first camera and the third camera.

[0057] S423: Input the vehicle exit distance or the vehicle entry distance into a relative retention distance calculation formula to obtain a relative retention distance of the vehicle.

[0058] The calculation formula of the relative distance of retention is as follows:

[0059]

[0060] Wherein, ln( ) represents a logarithmic function with a natural constant as the base; S is a distance constant, which is set to 1 in this embodiment to prevent the vehicle from entering at a distance of zero; ens represents the vehicle’s entry distance; It indicates the vehicle's exit distance; det indicates the relative distance of the vehicle in the community's vehicle observation frame, reflecting the impact of the vehicle on the community's vehicle congestion level.

[0061] In the above calculation formula (1), on the one hand, when the owner's vehicle A's driving distance is smaller, the probability of the vehicle stopping is greater, the time it takes to enter the parking lot is shorter, the possibility of causing vehicle congestion in the community is lower, and the corresponding calculated relative distance of vehicle detention is smaller, and the impact on the degree of vehicle congestion in the community is smaller. On the other hand, when the owner's vehicle A is farther away from the registered parking lot, the impact of the change in the corresponding vehicle's driving distance on the degree of vehicle congestion does not change much; and when the owner's vehicle A is farther away from the registered parking lot, the impact of the change in different vehicle driving distances on the time it takes for the vehicle to enter the parking lot and the degree of congestion is quite different. Therefore, the above calculation formula can reflect the changes in the impact of different vehicle driving distances on the degree of vehicle congestion;

[0062] Similarly, if the owner vehicle A does not appear at the entrance gate of the community within the previous L community vehicle observation intervals, it means that it is leaving the community and taking the nearest community exit gate as its destination. The relative distance of the vehicle stay of the owner vehicle A in the community vehicle observation frame is calculated in the same way using formula (2).

[0063] It should be noted that, in this embodiment, the relative vehicle detention distance of non-owner vehicles is uniformly set to 1.

[0064] S43. Calculate the relative distance of vehicle detention corresponding to each owner vehicle and the relative distance of vehicle detention corresponding to all non-owner vehicles in each frame of road image during the real-time observation period to obtain the dynamic density of vehicles in the community corresponding to each camera area during the real-time observation period, and the relative distance of vehicle detention corresponding to all non-owner vehicles is a preset value.

[0065] Among them, step S43 can be performed by the following steps.

[0066] S431, normalizing the relative vehicle detention distances corresponding to all owner vehicles, and updating the relative vehicle detention distances corresponding to all owner vehicles to the normalized results.

[0067] S432: Accumulate the relative vehicle detention distances corresponding to all owner vehicles and all non-owner vehicles corresponding to each frame of road image in the camera area to obtain the community vehicle density corresponding to each frame of road image.

[0068] The community vehicle density in this step reflects the vehicle density in this camera area and its impact on the degree of community vehicle congestion. When the relative distance of the owner's vehicle is smaller and the total number of vehicles in the camera area is smaller, the calculated community vehicle density in the camera area is smaller, and its impact on the degree of community vehicle congestion is smaller.

[0069] However, since the number of vehicles in the camera area is constantly changing during the observation time period, the vehicle density in the community is also changing. The community vehicle density corresponding to each frame of road image cannot reflect the changes in the impact of vehicles in the camera area during the entire community vehicle observation period on the degree of community vehicle congestion. Therefore, this embodiment also includes steps S433 to S437.

[0070] S433, summarizing the vehicle density of the residential area corresponding to each frame of the road image to obtain the change of the vehicle density of the residential area in each camera area within the observation time period.

[0071] S434. Perform mean calculation based on the cell vehicle density corresponding to each frame of the road image to obtain the mean cell vehicle density of each camera area within the observation time period.

[0072] S435. Perform straight line fitting on the change of the density of vehicles in the community of each camera area within the observation time period according to the time series relationship, and obtain the slope of the fitting straight line corresponding to each camera area.

[0073] S436: Normalize the slopes of the fitting lines corresponding to all the imaging areas to obtain a density coefficient corresponding to each imaging area.

[0074] S437. Calculate the mean value and density coefficient of the community vehicle density corresponding to each of the camera areas during the observation period to obtain the community vehicle dynamic density corresponding to each camera area.

[0075] The above-calculated dynamic density of community vehicles can reflect the impact of vehicles in a camera area on the degree of community vehicle congestion during the observation period. On the one hand, the mean value of the community vehicle density during the observation period represents the average level of the impact of vehicles in the camera area on the degree of community vehicle congestion. On the other hand, when the community vehicle density during the observation period shows an upward trend, the greater the slope of the fitting line, the greater the congestion impact of vehicles in the camera area on the subsequent community, and the greater the calculated dynamic density of community vehicles, which improves the utilization rate of community vehicle information and the sensitivity of reflecting the degree of community congestion.

[0076] S5, according to the evaluation speed calculation of all the vehicles corresponding to each camera area in each observation time period, the community traffic flow rate corresponding to each camera area in each observation time period is obtained, and the community traffic flow rate, community vehicle dynamic density and community traffic flow are recorded as the traffic flow feature vector in the observation time period;

[0077] In this embodiment, considering that the speed of vehicles in the community is slower than that on public roads, the existing vehicle flow rate calculation method is not suitable for the application environment of this embodiment. Therefore, in this embodiment, steps S51, S52, S53, S54, S55, S56 and S57 are included to calculate the community vehicle flow rate. Figure 5 , which shows a schematic flow chart of step S5 provided in an embodiment 1 of the present invention, including:

[0078] S51, summarizing all the vehicle evaluation speeds corresponding to each camera area to obtain a vehicle evaluation speed change of each vehicle within an observation time period.

[0079] S52. Calculate the average value of the vehicle evaluation speed of each vehicle during the observation period to obtain the average value of the vehicle evaluation speed corresponding to each vehicle.

[0080] S53. Perform a straight line fitting on the changes in the vehicle evaluation speed of each vehicle within the observation time period according to the time series relationship to obtain the slope of the fitting line corresponding to each vehicle.

[0081] S54. Normalize the slope of the fitting line corresponding to each vehicle to obtain a speed coefficient corresponding to each vehicle.

[0082] S55. Calculate the vehicle dynamic speed of each vehicle based on the speed coefficient corresponding to each vehicle and the average vehicle evaluation speed.

[0083] The vehicle dynamic rate of each vehicle obtained by calculation from step S51 to step S55 can reflect the changing trend of the vehicle's driving speed in an observation time period. On the one hand, the larger the mean of the vehicle evaluation speed, the faster the average driving speed of the vehicle, and the larger the vehicle dynamic rate, indicating that the congestion level in the camera area is low at this time. On the other hand, when the vehicle evaluation speed shows an upward trend in the vehicle observation interval of the community, the larger the slope of the fitting straight line, indicating that the vehicle's subsequent driving trend is accelerating, and the larger the calculated vehicle dynamic rate, reflecting that the traffic conditions in the camera area are good.

[0084] S56. According to the appearance relationship of the vehicles in each of the camera areas and the vehicle dynamic speed of each vehicle, respectively calculate the vehicle dynamic speed mean and the vehicle dynamic speed standard deviation corresponding to each of the camera areas.

[0085] S57. Calculate the traffic flow rate of the community in each of the camera areas according to the mean value of the vehicle dynamic rate and the standard deviation of the vehicle dynamic rate corresponding to each of the camera areas, and obtain the traffic flow rate of the community in each of the camera areas.

[0086] The community traffic flow rate obtained by calculation in step S56 and step S57 can reflect the vehicle speed in the entire camera area and the vehicle traffic conditions in the camera area within an observation time period. The larger the mean value of the vehicle dynamic rate, the faster the average speed of the vehicles in the camera area, and the overall driving trend is accelerating. The better the traffic conditions in the camera area, the larger the corresponding community traffic flow rate. At the same time, when there is a lane in the camera area that is unobstructed and another lane that is congested, the standard deviation of the vehicle dynamic rate is large. At this time, the traffic conditions in the camera area are poor, and the calculated community traffic flow rate is smaller, which improves the utilization rate of vehicle speed information and more accurately reflects the relative congestion level of vehicles in the camera area.

[0087] S6. Predictions are made based on the traffic flow feature vector corresponding to each observation time period and a preset prediction model to obtain predicted traffic flow feature vectors corresponding to each camera area.

[0088] Specifically, the prediction model preset in this step is a gated recurrent unit model (GRU model). The hidden layer size of the GRU model is 64, the learning rate is 0.01, the activation function uses the tanh function, and the optimizer uses the Adam optimizer. The specific construction and use method is the existing technology and will not be repeated in this embodiment.

[0089] Furthermore, in this embodiment, in order to obtain the change of vehicle congestion level in the community and guide the gate at the entrance of the community to release vehicles, the traffic conditions of the community are predicted by combining the hypergraph and the GRU model. Specifically, this step includes step S61, step S62 and step S63. Figure 6 , which shows a schematic flow chart of step S6 provided in an embodiment 1 of the present invention, including:

[0090] S61. Construct a hypergraph for each observation time period respectively to obtain a community vehicle hypergraph corresponding to each observation time period, wherein each node of the community vehicle hypergraph includes a traffic feature vector, and the spacing between all corresponding camera areas of the nodes of the community vehicle hypergraph and the nodes connected thereto is less than a preset distance.

[0091] It should be noted that in this embodiment, the probability of the same vehicle appearing repeatedly in adjacent camera areas is high, and their dynamic correlation is strong. Therefore, when constructing the hypergraph, a camera area is randomly selected, and the R camera areas with the closest distance to it are selected to form a hyperedge, and one is randomly selected from the R camera areas, and a new hyperedge is generated in the above manner until all N camera areas form corresponding hyperedges, completing the construction of the community traffic hypergraph, where the value of R in this solution is 5.

[0092] S62: Calculate the association matrix corresponding to each observation time period according to all the cell vehicle hypergraphs. It should be noted that the association matrix calculation method is the prior art and will not be described in detail in this embodiment.

[0093] S63, inputting a preset number of cell vehicle hypergraphs and corresponding association matrices into a preset prediction model for prediction, and obtaining a predicted traffic flow feature vector corresponding to each camera area.

[0094] It should be noted that in this implementation, the traffic feature vectors corresponding to multiple observation time periods are used to form a historical window for prediction. Specifically, in this embodiment, the real-time observation time period is taken as the last interval, and P-1 observation time periods are selected forward to form a community traffic history window. In this embodiment, the value of P is 8. The association matrix of the P community traffic hypergraphs in the community traffic history window and the traffic feature vectors of each node are input into the GRU model, and the association matrix of the P community traffic hypergraphs and the corresponding traffic feature vectors are output in turn, and the community vehicle observation intervals corresponding to the P community traffic hypergraphs are organized into a community traffic prediction window in chronological order.

[0095] In step S6, the traffic feature vector of each camera area is used as a node of the community traffic hypergraph, so as to enhance the utilization rate of community vehicle information, improve the accuracy of constructing the community traffic hypergraph, and further improve the accuracy of predicting the degree of community vehicle congestion.

[0096] S7. Calculate a community vehicle congestion index based on the predicted traffic flow feature vector corresponding to each camera area, and control the gate of the community according to the community vehicle congestion index.

[0097] Specifically, this step includes step S71, step S72 and step S73. Figure 7 , which shows a schematic flow chart of step S7 provided in an embodiment 1 of the present invention, including:

[0098] S71. Calculate the congestion index based on the congestion index calculation formula and the predicted vehicle flow feature vector corresponding to each camera area to obtain the vehicle congestion index corresponding to each camera area.

[0099] The congestion index calculation formula is as follows:

[0100]

[0101] in, represents the vehicle congestion index; Indicates the traffic flow in the community; Indicates the dynamic density of vehicles in the community; represents the traffic flow rate constant, which is set to 1 in this embodiment; Indicates the traffic velocity of the community.

[0102] In the above calculation formula, the vehicle congestion index reflects the degree of congestion in the camera area during the current observation period. The greater the community traffic flow and the community vehicle dynamic density, the smaller the community traffic rate, the greater the traffic flow in the camera area, the slower the speed, the higher the degree of vehicle congestion, and the greater the calculated community local vehicle congestion coefficient.

[0103] S72. Accumulate the vehicle congestion indexes corresponding to all the camera areas to obtain a community vehicle congestion index.

[0104] S73. Determine whether to allow vehicles to enter the community within each preset time period according to the community vehicle congestion index.

[0105] Among them, those skilled in the art can determine whether to allow a vehicle to enter the car by setting a threshold in the simplest way, thereby achieving the purpose of managing the vehicles in the community. At the same time, this embodiment also provides a method for managing community vehicles based on historical information. The details are as follows:

[0106] First, in step S63, the community vehicle observation intervals corresponding to P community traffic hypergraphs are obtained and organized into community traffic prediction windows in chronological order, so as to calculate the mean of the community vehicle congestion index within the community traffic prediction window, which is recorded as the mean of the community vehicle congestion index.

[0107] Secondly, since this method is in a state of continuous operation, the vehicle congestion index within the community traffic history window can be calculated according to step S71 and step S72.

[0108] Then, we can analyze the relationship between the mean value of the community vehicle congestion index and the vehicle congestion index in the community traffic history window. The specific analysis process is as follows:

[0109] When the mean value of the community vehicle congestion index does not exceed the median value of the community vehicle congestion index in the community traffic history window, the overall congestion level of the community is controllable, and the community entrance gate does not restrict vehicles from entering the community; when the mean value of the community vehicle congestion index in the community traffic prediction window is greater than the median value of the community vehicle congestion index in the community traffic history window, but does not exceed its third quartile, the community traffic volume is significantly increased, and the vehicle driving speed is significantly slowed down, indicating that a slight congestion has occurred in the community at this time, and the community entrance gate allows one vehicle to enter the community every Q community vehicle observation intervals, and in this embodiment, Q takes the value of 1; when the mean value of the community vehicle congestion index in the community traffic prediction window is greater than the third quartile of the community vehicle congestion index in the community traffic history window, it indicates that serious vehicle congestion has occurred in the community, and the community entrance gate prohibits vehicles from entering the community to prevent the congestion from further deteriorating.

[0110] In this embodiment, the YOLOv4 algorithm is used to process the collected video set to obtain the vehicle evaluation speed and community traffic flow. The vehicle density of the community is calculated using the owner's vehicle information and the positional relationship between the camera and the parking lot. According to the changing trend of the community vehicle density during the observation period, the dynamic density of the community vehicles is calculated to improve the utilization rate of the community vehicle information and the sensitivity of reflecting the degree of congestion in the community. According to the change of the vehicle evaluation speed during the observation period and the average driving speed of the vehicles in the camera area, the community traffic rate is calculated to improve the utilization rate of the vehicle speed information and more accurately reflect the relative congestion degree of the vehicles in the camera area. Finally, the optimized traffic flow parameters (vehicle evaluation speed, community traffic flow and community vehicle dynamic density) are used to construct the traffic feature vector, the traffic condition of the community is modeled as a community traffic hypergraph, and the GRU model is used to predict the traffic condition of the community afterwards, so as to improve the accuracy of the construction of the community traffic hypergraph, and thus improve the accuracy of the prediction of the degree of congestion of the community vehicles. According to the traffic feature vector of the community traffic hypergraph, the community vehicle congestion index is calculated. According to the changing trend of the community vehicle congestion index, different gate control strategies are adopted to complete the control of the release of vehicles entering the community, improve the accuracy of monitoring the degree of vehicle congestion in the community, reasonably control the vehicles entering the community, and reduce the occurrence of vehicle congestion in the community.

[0111] Embodiment 2:

[0112] like Figure 8 As shown, this embodiment provides a structural diagram of a community vehicle management system based on AI vehicle identification, and the system includes:

[0113] A camera module, used to obtain a video set of a preset duration, wherein the video set is obtained by shooting roads in different areas by at least three cameras, the preset duration is composed of a plurality of observation time periods, and the end time of the preset duration is the current time;

[0114] A classification module, used to label and classify the vehicles in the video set to obtain a labeled video set;

[0115] The first calculation module is used to calculate the traffic flow according to the preset camera acquisition frame rate and the label video set, and obtain the community traffic flow corresponding to each camera area and the vehicle evaluation speed of each vehicle in each observation time period;

[0116] The second calculation module is used to perform dynamic density calculation based on the preset camera layout information and label video set in the community to obtain the dynamic density of community vehicles corresponding to each camera area in each observation time period;

[0117] A third calculation module is used to calculate the community traffic rate corresponding to each camera area in each observation time period according to the evaluated speed of all the vehicles corresponding to each camera area in each observation time period, and record the community traffic rate, community vehicle dynamic density and community traffic flow as the traffic feature vector in the observation time period;

[0118] A prediction module is used to make predictions based on the traffic flow feature vector corresponding to each observation time period and a preset prediction model to obtain a predicted traffic flow feature vector corresponding to each camera area;

[0119] The gate control module is used to calculate the community vehicle congestion index based on the predicted traffic flow feature vector corresponding to each camera area, and control the gate of the community according to the community vehicle congestion index.

[0120] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0121] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0122] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A community vehicle management method based on AI vehicle recognition, It is characterized in that The method comprises: Acquire a video set of a preset duration, where the video set is obtained by shooting roads in different areas by at least three cameras, the preset duration is composed of a plurality of observation time periods, and the end time of the preset duration is the current time; Labeling and classifying the vehicles in the video set to obtain a labeled video set; The vehicle flow is calculated based on the preset camera acquisition frame rate and label video set to obtain the community traffic flow corresponding to each camera area and the vehicle evaluation speed of each vehicle in each observation time period; The dynamic density is calculated based on the preset camera layout information and label video set in the community, and the dynamic density of the community vehicles corresponding to each camera area in each observation time period is obtained; According to the evaluation speed calculation of all the vehicles corresponding to each camera area in each observation time period, the traffic flow rate of the community corresponding to each camera area in each observation time period is obtained, and the traffic flow rate of the community, the dynamic density of vehicles in the community and the traffic flow of the community are recorded as the traffic feature vector in the observation time period; Predictions are made based on the traffic flow feature vector corresponding to each observation time period and a preset prediction model to obtain a predicted traffic flow feature vector corresponding to each camera area; The vehicle congestion index of the community is calculated based on the predicted traffic flow feature vector corresponding to each camera area, and the gate of the community is controlled according to the vehicle congestion index of the community; The vehicles on each frame of the road image in the label video set are classified as owner vehicles or non-owner vehicles. The dynamic density is calculated according to the preset camera layout information in the community and the label video set to obtain the community vehicle dynamic density corresponding to each camera area in each observation time period, including: Respectively query the appearance status of the real-time owner vehicle in the historical label video set, the real-time owner vehicle is the owner vehicle that appears in the label video set corresponding to the real-time observation time period, the end time of the real-time observation time period is the current time, the historical label video set is the label video set corresponding to multiple historical observation time periods, and the historical observation time period is at least two observation time periods within a preset time length except the real-time observation time period; According to the preset camera layout information in the community, the appearance status of each owner's vehicle and the label video set, the relative vehicle detention distance corresponding to each owner's vehicle in each frame of road image in the real-time observation time period is obtained; The dynamic density of vehicles in the community corresponding to each camera area in the real-time observation period is obtained according to the relative distance of vehicle detention corresponding to each of the owner's vehicles and the relative distance of vehicle detention corresponding to all the non-owner vehicles in each frame of road image in the real-time observation period, and the relative distance of vehicle detention corresponding to all the non-owner vehicles is a preset value; The vehicle flow is calculated based on the preset camera acquisition frame rate and label video set to obtain the community traffic flow corresponding to each camera area in each observation time period and the vehicle evaluation speed of each vehicle, including: Count the total number of frames in which each vehicle appears in the camera area during the observation period to obtain the vehicle frame number; The vehicle evaluation speed of each vehicle in each camera area is obtained according to the duration of the observation period, the preset camera acquisition frame rate and the vehicle frame number; The total number of vehicles entering and leaving each camera area during the observation period is counted to obtain the traffic flow of the community corresponding to each camera area; According to the preset camera deployment information in the community, the appearance status of each owner's vehicle and the label video set, the relative vehicle detention distance corresponding to each owner's vehicle in each frame of road image in the real-time observation time period is obtained, including: If the appearance status of the owner's vehicle is "already appeared", the distance between the first camera and the second camera is extracted from the preset camera layout information in the community and recorded as the vehicle entry distance, where the first camera is the camera that photographs the owner's vehicle and the second camera is the camera located in the community parking lot; If the appearance status of the owner's vehicle is not present, the distance between the first camera and the third camera is extracted from the preset camera layout information in the community and recorded as the vehicle exit distance. The third camera is a camera located at the gate at the community exit. The vehicle exit distance or the vehicle entry distance is input into a relative retention distance calculation formula to obtain a relative retention distance of the vehicle.

2. The community vehicle management method based on AI vehicle identification according to claim 1, It is characterized in that The vehicles in the video set are labeled and classified to obtain a labeled video set, including: Identify the vehicle license plate in each frame of the video set according to a preset YOLOv4 mathematical model to obtain the license plate information corresponding to each vehicle; Send the license plate information corresponding to each vehicle to the community vehicle management system to obtain feedback information; The vehicles in each frame are classified and labeled according to the feedback information to obtain a labeled video set.

3. The community vehicle management method based on AI vehicle identification according to claim 1, It is characterized in that The dynamic density of vehicles in the cell corresponding to each camera area in the real-time observation time period is obtained according to the relative distance of vehicle detention corresponding to each of the owner's vehicles and the relative distance of vehicle detention corresponding to all the non-owner's vehicles in each frame of the road image in the real-time observation time period, including: Normalizing the relative vehicle detention distances corresponding to all owner vehicles, and updating the relative vehicle detention distances corresponding to all owner vehicles to the normalized results; The relative distances of vehicle detention corresponding to all owner vehicles and all non-owner vehicles corresponding to each frame of road image in the camera area are accumulated to obtain the community vehicle density corresponding to each frame of road image; Summarizing the cell vehicle density corresponding to each frame of the road image to obtain the cell vehicle density change of each camera area within the observation time period; Calculate the mean of the vehicle density of the cell corresponding to each frame of the road image to obtain the mean of the vehicle density of the cell in each camera area within the observation time period; Perform straight line fitting on the change of vehicle density in the community of each camera area within the observation time period according to the time series relationship, and obtain the slope of the fitting straight line corresponding to each camera area; Normalizing the slopes of the fitted lines corresponding to all the imaging areas to obtain a density coefficient corresponding to each imaging area; The mean value and density coefficient of the cell vehicle density corresponding to each of the camera areas during the observation period are calculated to obtain the cell vehicle dynamic density corresponding to each camera area.

4. The community vehicle management method based on AI vehicle identification according to claim 1, It is characterized in that According to the calculation of all the vehicle evaluation speeds corresponding to each camera area in each observation time period, the community traffic flow rate corresponding to each camera area in each observation time period is obtained, including: Summarizing all the vehicle evaluation speeds corresponding to each camera area to obtain a vehicle evaluation speed change of each vehicle during an observation period; The average value of the vehicle evaluation speed change of each vehicle during the observation period is calculated to obtain the average vehicle evaluation speed corresponding to each vehicle; According to the time series relationship, a straight line fitting is performed on the changes in the vehicle evaluation speed of each vehicle during the observation period to obtain the slope of the fitting line corresponding to each vehicle; The slope of the fitting line corresponding to each vehicle is normalized to obtain the speed coefficient corresponding to each vehicle; The vehicle dynamic speed of each vehicle is obtained by calculating the speed coefficient corresponding to each vehicle and the average vehicle evaluation speed; According to the appearance relationship of the vehicles in each of the camera areas and the vehicle dynamic speed of each vehicle, respectively calculating the vehicle dynamic speed mean and the vehicle dynamic speed standard deviation corresponding to each of the camera areas; The traffic flow rate of the community in each camera area is obtained by calculating the mean value of the vehicle dynamic rate and the standard deviation of the vehicle dynamic rate corresponding to each camera area.

5. The community vehicle management method based on AI vehicle identification according to claim 1, It is characterized in that According to the traffic flow feature vector corresponding to each observation time period and the preset prediction model, the predicted traffic flow feature vector corresponding to each camera area is obtained, including: Constructing a hypergraph for each observation time period respectively to obtain a cell vehicle hypergraph corresponding to each observation time period, wherein each node of the cell vehicle hypergraph includes a vehicle flow feature vector, and the spacing between all corresponding camera areas of the cell vehicle hypergraph and the nodes connected thereto is less than a preset distance; The association matrix corresponding to each observation time period is calculated based on all the cell vehicle hypergraphs; According to the preset number of cell vehicle hypergraphs and the corresponding association matrix, a preset prediction model is input for prediction to obtain the predicted traffic flow feature vector corresponding to each camera area.

6. The community vehicle management method based on AI vehicle identification according to claim 1, It is characterized in that The vehicle congestion index of the community is calculated based on the predicted traffic flow feature vector corresponding to each camera area, and the gate of the community is controlled according to the vehicle congestion index of the community, including: The congestion index is calculated based on the congestion index calculation formula and the predicted traffic flow feature vector corresponding to each camera area, and the vehicle congestion index corresponding to each camera area is obtained; Accumulating the vehicle congestion indexes corresponding to all the camera areas to obtain a community vehicle congestion index; Whether to allow vehicles to enter the community within each preset time period is determined according to the community vehicle congestion index.

7. A community vehicle management system based on AI vehicle identification, It is characterized in that include: A camera module, used to obtain a video set of a preset duration, wherein the video set is obtained by shooting roads in different areas by at least three cameras, the preset duration is composed of a plurality of observation time periods, and the end time of the preset duration is the current time; A classification module, used to label and classify the vehicles in the video set to obtain a labeled video set; The first calculation module is used to calculate the traffic flow according to the preset camera acquisition frame rate and the label video set, and obtain the community traffic flow corresponding to each camera area and the vehicle evaluation speed of each vehicle in each observation time period; The second calculation module is used to perform dynamic density calculation based on the preset camera layout information and label video set in the community to obtain the dynamic density of community vehicles corresponding to each camera area in each observation time period; A third calculation module is used to calculate the community traffic rate corresponding to each camera area in each observation time period according to the evaluated speed of all the vehicles corresponding to each camera area in each observation time period, and record the community traffic rate, community vehicle dynamic density and community traffic flow as the traffic feature vector in the observation time period; A prediction module is used to make predictions based on the traffic flow feature vector corresponding to each observation time period and a preset prediction model to obtain a predicted traffic flow feature vector corresponding to each camera area; A gate control module, used to calculate a community vehicle congestion index based on the predicted traffic flow feature vector corresponding to each camera area, and control the gate of the community according to the community vehicle congestion index; The vehicles on each frame of the road image in the label video set are classified as owner vehicles or non-owner vehicles. The dynamic density is calculated according to the preset camera layout information in the community and the label video set to obtain the community vehicle dynamic density corresponding to each camera area in each observation time period, including: Respectively query the appearance status of the real-time owner vehicle in the historical label video set, the real-time owner vehicle is the owner vehicle that appears in the label video set corresponding to the real-time observation time period, the end time of the real-time observation time period is the current time, the historical label video set is the label video set corresponding to multiple historical observation time periods, and the historical observation time period is at least two observation time periods within a preset time length except the real-time observation time period; According to the preset camera layout information in the community, the appearance status of each owner's vehicle and the label video set, the relative vehicle detention distance corresponding to each owner's vehicle in each frame of road image in the real-time observation time period is obtained; The dynamic density of vehicles in the community corresponding to each camera area in the real-time observation period is obtained according to the relative distance of vehicle detention corresponding to each of the owner's vehicles and the relative distance of vehicle detention corresponding to all the non-owner vehicles in each frame of road image in the real-time observation period, and the relative distance of vehicle detention corresponding to all the non-owner vehicles is a preset value; The vehicle flow is calculated based on the preset camera acquisition frame rate and label video set to obtain the community traffic flow corresponding to each camera area in each observation time period and the vehicle evaluation speed of each vehicle, including: Count the total number of frames in which each vehicle appears in the camera area during the observation period to obtain the vehicle frame number; The vehicle evaluation speed of each vehicle in each camera area is obtained according to the duration of the observation period, the preset camera acquisition frame rate and the vehicle frame number; The total number of vehicles entering and leaving each camera area during the observation period is counted to obtain the traffic flow of the community corresponding to each camera area; According to the preset camera deployment information in the community, the appearance status of each owner's vehicle and the label video set, the relative vehicle detention distance corresponding to each owner's vehicle in each frame of road image in the real-time observation time period is obtained, including: If the appearance status of the owner's vehicle is "already appeared", the distance between the first camera and the second camera is extracted from the preset camera layout information in the community and recorded as the vehicle entry distance, where the first camera is the camera that photographs the owner's vehicle and the second camera is the camera located in the community parking lot; If the appearance status of the owner's vehicle is not present, the distance between the first camera and the third camera is extracted from the preset camera layout information in the community and recorded as the vehicle exit distance. The third camera is a camera located at the gate at the community exit. The vehicle exit distance or the vehicle entry distance is input into a relative retention distance calculation formula to obtain a relative retention distance of the vehicle.