Smart park vehicle positioning and tracking system and method based on video surveillance

Abstract

The present invention relates to the field of vehicle tracking and positioning, and in particular to a vehicle positioning and tracking system and method for a smart park based on video surveillance. The system integrates dynamic grid division, precise vehicle positioning, intelligent path planning and user interaction functions, and realizes intelligent division and smoothness evaluation of the park area through a dynamic grid module. The vehicle positioning module uses a deep learning algorithm to realize efficient identification and tracking of vehicles, and the path planning module builds a dynamic path library based on real-time road conditions to recommend the optimal driving path for the vehicle. The interactive module not only supports rapid model training and deployment, but also realizes the visual display of processing results, greatly improving the intelligent level of park vehicle management; the present invention effectively solves the problems of inaccurate positioning and unreasonable path planning in smart park vehicle management, and improves the traffic efficiency and management effectiveness of the park.

Description

Technical Field

[0001] The present invention relates to the field of vehicle tracking and positioning, and in particular to a smart park vehicle positioning and tracking system and method based on video surveillance. Background Art

[0002] Existing smart parks integrate advanced technologies such as the Internet of Things, big data, cloud computing and artificial intelligence to achieve intelligent management and services in the parks, thereby improving the safety, convenience and efficiency of the parks. Among them, the vehicle positioning and tracking system is an indispensable part of the smart park. It is responsible for monitoring the real-time location of vehicles in the park, providing data support for park management, and also providing navigation services for vehicles to improve traffic smoothness and safety.

[0003] For example, a Chinese patent with authorization announcement number CN117635663B discloses a target vehicle video tracking method and electronic equipment, which obtains the real-time position of the target vehicle and converts it into the system coordinates of the road model, then calculates the real-time distance between the target vehicle and each camera, and displays the video shot by the target camera whose real-time distance is less than the preset distance in the display interface.

[0004] For example, a Chinese patent application with publication number CN113516688A discloses a multi-dimensional intelligent positioning and tracking system for vehicles, including a video acquisition module installed on a vehicle for capturing road images and scene images during vehicle driving. The multi-dimensional intelligent positioning and tracking system obtains a first feature point based on adjacent frames of scene images, calculates the spatial position of the first feature point in combination with the working state of the acquisition device, calculates the spatial position of a second feature point between a subsequent frame of scene image and a next frame of scene image using the spatial position of the first feature point, obtains first vehicle position information based on the spatial position of the second feature point, obtains second vehicle position information based on the road image and the scene image, and comprehensively determines the vehicle position in combination with the first vehicle position information and the second vehicle position information.

[0005] The above-mentioned prior art has the following problems: due to the regional limitations of the park, it is difficult for the prior art to make accurate planning and allocation guidance for the paths of each type of vehicle in the park, resulting in congestion in the park or the allocated paths being impassable. In order to solve the above-mentioned problems, the present invention provides a smart park vehicle positioning and tracking system and method based on video surveillance. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a smart park vehicle positioning and tracking system and method based on video surveillance. The system integrates dynamic grid division, precise vehicle positioning, intelligent path planning and user interaction functions. The dynamic grid module is used to realize the intelligent division and smoothness assessment of the park area. The vehicle positioning module uses a deep learning algorithm to realize efficient identification and tracking of vehicles. The path planning module builds a dynamic path library based on real-time road conditions to recommend the optimal driving path for the vehicle. The interactive module not only supports rapid model training and deployment, but also realizes the visual display of processing results, which greatly improves the level of intelligent park vehicle management. The present invention effectively solves the problems of inaccurate positioning and unreasonable path planning in smart park vehicle management, and improves the traffic efficiency and management effectiveness of the park.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] Smart park vehicle positioning and tracking system based on video surveillance, including: dynamic grid module, vehicle positioning module and path planning module;

[0009] The dynamic grid module includes a data acquisition unit, an area division unit, an area evaluation unit and an area labeling unit; the data acquisition unit is used to collect the work park area map and the park real attribute data, and pre-process and save them; the area division unit is used to divide the area through a grid algorithm according to the acquired park area map and the park real attribute data; the area evaluation unit is used to evaluate the smoothness of the divided corresponding area through an evaluation algorithm according to the traffic flow data at different time points in the park real attribute data, and obtain the smoothness of the corresponding area at different time points; the area labeling unit is used to dynamically classify and label the driving paths in the divided area according to the smoothness of the corresponding area at different time points, and update the labeled paths to the path library constructed by the path planning unit;

[0010] The vehicle positioning module includes a vehicle classification unit and a tracking and positioning unit; the vehicle classification unit is used to detect and classify vehicles based on the vehicle video data captured by the monitoring in real time within the divided area through the built-in multi-dimensional detection and matching model; the tracking and positioning unit is used to track and locate the classified vehicles in real time through the built-in multi-target tracking and positioning algorithm;

[0011] The path planning module includes a path planning unit and a path recommendation unit; the path planning unit is used to construct a path library, and use the path optimization model built into the path library to perform secondary planning adjustments on the constructed path according to the real attribute data of the park, the smoothness of the corresponding area at different time points and the detected vehicle type; the path recommendation unit is used to recommend the optimal path to the currently tracked and located vehicle through the constructed path library according to the current position of the vehicle, the vehicle detection classification result, the target point position and the real attribute data of the park at the current moment;

[0012] The specific steps of constructing the path optimization model include:

[0013] D1. Based on the acquired park's real attribute data, the divided park areas, and the vehicle types and number of each type of vehicles in the corresponding divided areas obtained by the vehicle classification unit, a path optimization function is constructed to minimize the total time for each type of vehicle to pass through all divided areas. The specific objective function is:

[0014] ,

[0015] in, represents the decision variable for the j-th type of vehicle to use the i-th path in the n-th partition area, represents the occupied capacity of the i-th path of the j-th type of vehicle in the n-th partition area, Indicates the total number of areas divided into the park. Represents the total number of driving paths in all divided areas. Indicates the total number of vehicle types detected by monitoring in each divided area; represents the road capacity of the ith travel path in the nth partition area, It represents the mean value of the smoothness coefficient of the i-th traffic path of the j-th vehicle type in the n-th divided area within the next 5 minutes. represents the smoothness coefficient threshold, It represents the smoothness coefficient of the j-th vehicle traveling on the v-th traffic path in the u-th divided area at the current moment, The real-time travel time of the j-th type of vehicle on the i-th travel path in the n-th area;

[0016] D2. Build a path optimization model based on BP neural network optimization particle swarm algorithm, and build the constructed path optimization function into the particle swarm algorithm in the path optimization model as the fitness function of the particle swarm algorithm;

[0017] D3. Input the real attribute data of the park obtained in real time by video surveillance and the dynamically annotated path data in the existing path library into the path optimization model, obtain the optimized path library, and feed the optimized path library back to the path recommendation unit, so that each type of vehicle can reach the target point in the shortest time.

[0018] Specifically, the real attribute data of the park includes: path distribution data within the park, building and impassable area data, vehicle flow video data, video surveillance distribution data, single video surveillance coverage and target point location; the specific steps of the regional assessment unit to obtain the smoothness of the corresponding area at different time points include:

[0019] B1. Obtain the road capacity of the ith traffic path in the nth divided area within the three-month historical working time of the park , historical traffic volume of the jth type of vehicles , the average travel time of the j-th type of vehicle on the i-th travel path and path length , the average speed of the j-th type of vehicle when traveling smoothly on the i-th traffic path , and pre-process the acquired data;

[0020] B2, dividing the pre-processed data obtained in B1 into windows of 10 minutes in length, and obtaining the divided road driving status data;

[0021] B3. Using the divided road driving status data, calculate the flow density ratio corresponding to the current density and the blocking density and the driving speed ratio corresponding to the current speed and the free-flowing speed in each divided time window, specifically: , ,in, represents the travel speed ratio of the jth vehicle traveling on the i-th traffic path in the n-th divided area, represents the flow density ratio of the jth type of vehicle traveling on the i-th traffic path in the n-th divided area, represents the blocking density on the ith path in the nth partition area, represents the body width of the j-th vehicle, Represents the average width of the i-th path in the n-th partition area.

[0022] Specifically, the specific steps of the regional evaluation unit obtaining the smoothness of the corresponding region at different time points also include:

[0023] B4. Based on the calculated driving speed ratio and flow density ratio, the smoothness coefficient of the j-th vehicle traveling on the i-th traffic path in the n-th divided area is calculated. , and align the calculated smooth coefficient with the driving status data collected in B1 in the time dimension to obtain a smooth coefficient prediction training set;

[0024] B5. Construct a smooth traffic prediction model based on a support vector machine, input the obtained smooth traffic coefficient prediction training set into the smooth traffic prediction model for training, obtain the trained smooth traffic prediction model, and embed the obtained smooth traffic prediction model into the regional evaluation unit;

[0025] B6. Obtain the real attribute data and vehicle detection classification result data in the park in real time through video surveillance, and input the acquired data into the regional evaluation unit to calculate the average value of the smooth coefficient of the i-th traffic path of the j-th vehicle type in the n-th divided area within the next 5 minutes, and use the average value as the smooth coefficient of the j-th vehicle on the i-th driving path at the current moment. ;

[0026] B7. Based on the calculated smoothness coefficient of the i-th path of the j-th vehicle in the n-th divided area at the current moment, the length of the i-th path , calculate the real-time travel time of the j-th type of vehicle on the i-th travel path in the n-th area The calculated smoothness coefficient and real-time travel time are fed back to the regional annotation unit to dynamically update and annotate the corresponding regional driving path in the path library constructed by the path planning unit. Specifically, .

[0027] Specifically, the specific steps of the path recommendation unit recommending the optimal path include:

[0028] C1. Set the threshold of the smoothness coefficient and , the current smoothness coefficient of the i-th traffic path of the j-th vehicle in the n-th partition area is obtained and and For comparison, When , the j-th vehicle passes through the i-th traffic path in the n-th divided area;

[0029] C2. When When , we can calculate and Compare the travel time of the corresponding travel path with the current time smoothness coefficient and travel time corresponding to the travel paths in the remaining divided areas in the path library;

[0030] C21, when all comparison results meet and When , the j-th vehicle still travels along the i-th path in the n-th divided area until it reaches the target point, where It represents the smoothness coefficient of the j-th vehicle traveling on the v-th traffic path in the u-th divided area at the current moment, represents the real-time travel time of the j-th type of vehicle on the v-th travel path in the u-th area;

[0031] C22, when at least one of the paths in the divided area satisfies When and satisfaction The travel time corresponding to the conditional travel path When there is only one pass path that satisfies When The corresponding travel path is used as the regional travel path at the current time, and the C22 process is repeated until the target point is reached;

[0032] C23, when at least two paths satisfy When The travel paths are sorted in descending order according to the travel time, and the travel path with the smallest travel time after the descending order is selected as the regional driving path at the current time, and the C23 process is repeated until the target point is reached;

[0033] C24, when all the paths satisfy When , the j-th vehicle still travels along the i-th path in the n-th divided area until it reaches the target point.

[0034] Specifically, the specific steps of the path recommendation unit recommending the optimal path also include:

[0035] C3. When When the path planning unit is called, the path in the path library is optimized through the path optimization model according to the current vehicle type, vehicle position, target point position and the real attribute data of the park, and the mean value of the smooth coefficient and the real-time travel time within 5 minutes corresponding to the driving path in the corresponding area after optimization are obtained. ;

[0036] C4. Repeat C1 and C2 processes according to the optimized path library to recommend the corresponding regional driving path to the j-th type of vehicle identified by monitoring until it reaches the target point.

[0037] Specifically, the specific steps of the vehicle classification unit performing vehicle detection through the multi-dimensional detection matching model include:

[0038] E1. Perform frame processing on the vehicle flow video data in the acquired real attribute data of the park, obtain P types of vehicle image data, and annotate the corresponding license plate number, vehicle color and vehicle attribute information data on the framed vehicle image data;

[0039] E2. Build a multi-dimensional detection and matching model based on the pre-trained lightweight model, input the acquired vehicle flow video data into the lightweight VIT pre-trained sub-model in the multi-dimensional detection and matching model, extract the vehicle running status and vehicle-to-vehicle distance, and obtain the vehicle running status characteristics and vehicle-to-vehicle distance;

[0040] E3. Set the vehicle distance threshold, input the labeled P types of vehicle image data into the resnet-18 pre-trained sub-model at the same time, extract the static representation corresponding to each of the P types of vehicles, and obtain the static feature set corresponding to the P types of vehicles. ;

[0041] E4. Compare the obtained vehicle distance with the vehicle distance threshold. and When the inter-vehicle distance of the corresponding type of vehicles is greater than the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle can be distinguished. and When the inter-vehicle distance of the corresponding type of vehicles is less than or equal to the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle cannot be distinguished;

[0042] E5. Set vehicle similarity threshold , the obtained indistinguishable vehicle static features and Input into the similarity sub-model to calculate the similarity between the static features of the vehicle , and the calculated similarity Compared with the vehicle similarity threshold, when Less than When and The static features of the dissimilar vehicles are concatenated with the static features of distinguishable vehicles to obtain secondary distinguishable vehicle static features;

[0043] E6. When Greater than or equal to When and For similar vehicle static features, the obtained similar vehicle static features are cascaded to obtain a secondary indistinguishable vehicle static feature set.

[0044] Specifically, the specific steps of the vehicle classification unit performing vehicle detection through the multi-dimensional detection matching model also include:

[0045] E7. Input the vehicle attribute information data into the Chinese pre-trained Bert sub-model to obtain the corresponding vehicle text representation features;

[0046] E8. Input the acquired text representation features, the secondary indistinguishable vehicle static feature set and the vehicle running state features into the fusion attention sub-model constructed by the cross attention network to obtain the corresponding vehicle fusion distinguishing features;

[0047] E9. Set the cross entropy loss function and training cycle, input the obtained vehicle fusion distinguishing features and secondary distinguishable vehicle static features into the output sub-model constructed by the softmax function for training, obtain the trained multi-dimensional detection and matching model, and output the corresponding vehicle type, vehicle width, vehicle attribute information and the number of each type of vehicle in the corresponding divided area;

[0048] E10. The trained multi-dimensional detection and matching model is built into the path optimization model. The traffic flow video data in each divided area is captured in real time through video surveillance. The vehicle classification unit is used to identify and detect the vehicle type in the traffic flow video. The detected vehicle type and vehicle attribute information are fed back to the tracking and positioning unit and the path recommendation unit for positioning tracking and path recommendation.

[0049] The method for positioning and tracking vehicles in a smart park based on video surveillance includes the following steps:

[0050] S1. Obtain the park area map and the park's real attribute data and pre-process them, and use the grid algorithm to divide the area according to the saved data. At the same time, use the evaluation algorithm to obtain the smoothness of each area at different times after the division;

[0051] S2. Dynamically classify and label each divided area according to the obtained smoothness of each area at different times;

[0052] S3. According to the video data obtained from the corresponding divided areas, the vehicles are detected and classified through the built-in multi-dimensional detection and matching model, and each type of classified vehicle is tracked and located in real time through the multi-target tracking and positioning algorithm;

[0053] S4. Construct a path library with a path optimization model, and recommend existing or optimized travel paths to the corresponding tracking vehicles through a recommendation algorithm and the path library based on the current position of the corresponding tracking vehicle, the target point position and the real attribute data of the park at the current moment.

[0054] A computer-readable storage medium stores computer instructions, which, when executed, execute a method for positioning and tracking vehicles in a smart park based on video surveillance.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. In view of the deficiencies of the prior art, the present invention uses deep learning of historical data and real-time monitoring information to realize the intelligent division and smoothness assessment of the park map in the dynamic grid module, thus ensuring the accuracy and timeliness of path planning. The vehicle positioning module uses advanced multi-dimensional detection and matching models and multi-target tracking and positioning algorithms to accurately identify and track each vehicle, providing key real-time location information for path planning. In addition, the present invention also uses a path optimization model to dynamically adjust the path to adapt to changing traffic conditions. This mechanism can intelligently recommend the optimal path based on the vehicle type, current location, target point location, and real-time road conditions, and the smoothness and vehicle type data of each divided area’s driving path calculated in real time, effectively avoiding congestion, ensuring the feasibility of the path, and avoiding the infeasibility of the recommended path;

[0057] 2. The present invention deeply analyzes historical data, combines real-time monitoring information, and uses support vector machine (SVM) and BP neural network to optimize particle swarm algorithm to build a dynamic smooth prediction and path optimization model, which can not only evaluate the traffic conditions in various areas of the park in real time, but also provide personalized and efficient path planning for different types of vehicles, effectively avoiding congestion and invalid path allocation problems. In addition, the present invention can accurately identify vehicle type, size and vehicle attribute information through a multi-dimensional detection and matching model, further improving the accuracy of vehicle identification and positioning, so that the path assigned to each type of vehicle is more in line with its corresponding driving state and reaches the corresponding target point faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a module diagram of the smart park vehicle positioning and tracking system based on video surveillance of the present invention;

[0059] Figure 2 This is a working diagram of the smart park vehicle positioning and tracking system based on video surveillance of the present invention;

[0060] Figure 3 This is a flow chart of the method for positioning and tracking vehicles in a smart park based on video surveillance according to the present invention. DETAILED DESCRIPTION

[0061] Example 1

[0062] See also Figure 1 , an embodiment provided by the present invention: a smart park vehicle positioning and tracking system based on video surveillance, comprising: a dynamic grid module, a vehicle positioning module, a path planning module and an interaction module;

[0063] The dynamic grid module is used to dynamically grid the park and dynamically mark the areas according to the vehicle data of each area;

[0064] The dynamic grid module includes a data acquisition unit, an area division unit, an area evaluation unit and an area marking unit; the data acquisition unit is used to collect the work park area map and the park's real attribute data, and pre-process and save them; the park's real attribute data includes: path distribution data within the park, building and impassable area data, vehicle flow video data, video surveillance distribution data, single video surveillance coverage and target point location;

[0065] The area division unit is used to perform area division by a grid algorithm according to the acquired park area map and the real attribute data of the park; further, the specific steps implemented by the area division unit in this embodiment include:

[0066] A1. Receive the pre-processed park area map and the park's real attribute data from the data acquisition unit;

[0067] A2. Input the acquired data into the clustering algorithm to obtain the clustering points of the park monitoring distribution and the vehicle distribution clustering points. According to the clustering points, calculate the average monitoring vehicle data of a single monitoring in the corresponding clustering point, and use the inverse of the average monitoring vehicle data of a single monitoring as the regional division factor;

[0068] A3. Input the obtained regional division factors, park real attribute data, park monitoring distribution cluster points, vehicle distribution cluster points and park area map data into the grid algorithm to obtain N divided areas. The larger the average monitoring vehicle data of a single monitor, the smaller the corresponding divided area, and vice versa, so that areas with heavy traffic can be better monitored.

[0069] This process uses clustering algorithms to identify hot spots of monitoring, paths and vehicle distribution within the park. By calculating the average number of vehicle monitoring at each monitoring point and its inverse as the regional division factor, the monitoring demand density of different areas can be determined more scientifically, so that areas with dense traffic will be subdivided into smaller monitoring units, thereby ensuring that key areas receive more frequent monitoring coverage.

[0070] The regional evaluation unit is used to evaluate the smoothness of the divided corresponding areas through an evaluation algorithm according to the traffic flow data at different time points in the real attribute data of the park, and obtain the smoothness of the corresponding areas at different time points; further, in this embodiment, the specific steps of the regional evaluation unit obtaining the smoothness of the corresponding areas at different time points include:

[0071] B1. Obtain the road capacity of the ith traffic path in the nth divided area within the three-month historical working time of the park , historical traffic volume of the jth type of vehicles , the average travel time of the j-th type of vehicle on the i-th travel path and path length , the average speed of the j-th type of vehicle when traveling smoothly on the i-th traffic path , and pre-process the acquired data;

[0072] B2, dividing the pre-processed data obtained in B1 into windows of 10 minutes in length, and obtaining the divided road driving status data;

[0073] B3. Using the divided road driving status data, calculate the flow density ratio corresponding to the current density and the blocking density and the driving speed ratio corresponding to the current speed and the free-flowing speed in each divided time window, specifically: , ,in, represents the travel speed ratio of the jth vehicle traveling on the i-th traffic path in the n-th divided area, represents the flow density ratio of the jth type of vehicle traveling on the i-th traffic path in the n-th divided area, represents the blocking density on the ith path in the nth partition area, represents the body width of the j-th vehicle, represents the average width of the ith path in the nth partition area;

[0074] B4. Based on the calculated driving speed ratio and flow density ratio, the smoothness coefficient of the j-th vehicle traveling on the i-th traffic path in the n-th divided area is calculated. , and align the calculated smooth coefficient with the driving status data collected in B1 in the time dimension to obtain a smooth coefficient prediction training set;

[0075] B5. Construct a smooth traffic prediction model based on a support vector machine, input the obtained smooth traffic coefficient prediction training set into the smooth traffic prediction model for training, obtain the trained smooth traffic prediction model, and embed the obtained smooth traffic prediction model into the regional evaluation unit;

[0076] B6. Obtain the real attribute data and vehicle detection classification result data in the park in real time through video surveillance, and input the acquired data into the regional evaluation unit to calculate the average value of the smooth coefficient of the i-th traffic path of the j-th vehicle type in the n-th divided area within the next 5 minutes, and use the average value as the smooth coefficient of the j-th vehicle on the i-th driving path at the current moment. ;

[0077] B7. Based on the calculated smoothness coefficient of the i-th path of the j-th vehicle in the n-th divided area at the current moment, the length of the i-th path , calculate the real-time travel time of the j-th type of vehicle on the i-th travel path in the n-th area The calculated smoothness coefficient and real-time travel time are fed back to the regional annotation unit to dynamically update and annotate the corresponding regional driving path in the path library constructed by the path planning unit. Specifically, .

[0078] This process collects and analyzes traffic flow data during working hours in the past three months, and combines key indicators such as vehicle type, road capacity, average travel time and speed to build a database that fully reflects the actual operation status of the road. The data preprocessing and time window division in this process ensure the timeliness and accuracy of the analysis. Secondly, through in-depth mining of road driving status data, the flow density ratio and driving speed ratio are calculated, and then the smooth coefficient is obtained. This coefficient intuitively reflects the degree of traffic congestion in a specific area and time point. By aligning the smoothness coefficient with the historical driving status data in the time dimension to form a prediction training set, the smoothness prediction model established using the support vector machine (SVM) can be trained and optimized, enabling the system to make accurate predictions about future traffic conditions based on past data. Finally, the system can process video surveillance data in real time, and dynamically calculate the average smoothness coefficient within the next 5 minutes based on vehicle detection and classification results. This forward-looking function enables park managers to adjust traffic control measures in a timely manner to avoid potential congestion. The real-time travel time calculation and feedback mechanism not only optimizes the path library of the path planning unit, but also realizes the dynamic update and annotation of the driving path, ensuring that the driver can choose the best route based on the latest traffic conditions, reducing invalid waiting time and improving travel efficiency.

[0079] The area labeling unit is used to dynamically classify and label the driving paths in the divided areas according to the acquired smoothness of the corresponding areas at different time points, and update the labeled paths to the path library constructed by the path planning unit;

[0080] The vehicle positioning module is used to monitor, classify and track and locate the vehicle through a deep learning algorithm based on the acquired video data; the vehicle positioning module includes a vehicle classification unit and a tracking and positioning unit;

[0081] The vehicle classification unit is used to detect and classify vehicles according to the vehicle video data captured by the monitoring in real time within the divided area through the built-in multi-dimensional detection and matching model; further, in this embodiment, the specific steps of the vehicle classification unit performing vehicle detection through the multi-dimensional detection and matching model include:

[0082] E1. Perform frame processing on the vehicle flow video data in the acquired real attribute data of the park, obtain P types of vehicle image data, and annotate the vehicle image data after frame division with corresponding license plate number, vehicle color and vehicle attribute information data; further, in this embodiment, the vehicle attribute information data annotation includes vehicle type (such as car, truck, motorcycle), vehicle brand, vehicle model, vehicle size, whether it is loaded with goods and vehicle status;

[0083] E2. Build a multi-dimensional detection and matching model based on the pre-trained lightweight model, input the acquired vehicle flow video data into the lightweight VIT pre-trained sub-model in the multi-dimensional detection and matching model, extract the vehicle running status and vehicle-to-vehicle distance, and obtain the vehicle running status characteristics and vehicle-to-vehicle distance;

[0084] E3. Set the vehicle distance threshold, input the labeled P types of vehicle image data into the resnet-18 pre-trained sub-model at the same time, extract the static representation corresponding to each of the P types of vehicles, and obtain the static feature set corresponding to the P types of vehicles. ;

[0085] E4. Compare the obtained vehicle distance with the vehicle distance threshold. and When the inter-vehicle distance of the corresponding type of vehicles is greater than the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle can be distinguished. and When the inter-vehicle distance of the corresponding type of vehicles is less than or equal to the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle cannot be distinguished;

[0086] E5. Set vehicle similarity threshold , the obtained indistinguishable vehicle static features and Input into the similarity sub-model to calculate the similarity between the static features of the vehicle , and the calculated similarity Compared with the vehicle similarity threshold, when Less than When and The static features of the dissimilar vehicles are concatenated with the static features of distinguishable vehicles to obtain secondary distinguishable vehicle static features;

[0087] E6. When Greater than or equal to When and For similar vehicle static features, the obtained similar vehicle static features are cascaded to obtain a secondary indistinguishable vehicle static feature set.

[0088] E7. Input the vehicle attribute information data into the Chinese pre-trained Bert sub-model to obtain the corresponding vehicle text representation features;

[0089] E8. Input the acquired text representation features, the secondary indistinguishable vehicle static feature set and the vehicle running state features into the fusion attention sub-model constructed by the cross attention network to obtain the corresponding vehicle fusion distinguishing features;

[0090] E9. Set the cross entropy loss function and training cycle, input the obtained vehicle fusion distinguishing features and secondary distinguishable vehicle static features into the output sub-model constructed by the softmax function for training, obtain the trained multi-dimensional detection and matching model, and output the corresponding vehicle type, vehicle width, vehicle attribute information and the number of each type of vehicle in the corresponding divided area;

[0091] E10. The trained multi-dimensional detection and matching model is built into the path optimization model. The traffic flow video data in each divided area is captured in real time through video surveillance. The vehicle classification unit is used to identify and detect the vehicle type in the traffic flow video. The detected vehicle type and vehicle attribute information are fed back to the tracking and positioning unit and the path recommendation unit for positioning tracking and path recommendation.

[0092] Through the integration of deep learning and computer vision technology, the model can accurately perform frame processing and feature extraction on traffic flow videos. It can not only identify static attributes such as vehicle type, brand, and size, but also gain insight into dynamic information such as vehicle operating status and vehicle-to-vehicle distance. The setting of vehicle-to-vehicle distance threshold and vehicle similarity threshold further enhances the model's ability to distinguish similar vehicles and ensures the accuracy of recognition. The introduction of the Chinese pre-trained Bert model makes it possible to integrate text representation features with visual features, which strengthens the model's understanding of vehicle attribute information. With the help of the cross-attention network, the model can capture the feature differences of vehicles from multiple angles, and the output vehicle fusion distinction features are more comprehensive. After being built into the path optimization unit, the model can analyze traffic flow videos in various areas of the park in real time, quickly identify vehicle types and attributes, and provide strong support for vehicle positioning and tracking and path recommendation, greatly improving the smoothness and safety of park traffic.

[0093] The tracking and positioning unit is used to track and position the classified vehicles in real time through a built-in multi-target tracking and positioning algorithm; further, the multi-target tracking and positioning algorithm in this embodiment includes DeepSORT, FairMOT and DAMNet;

[0094] The path planning module is used to build a dynamic path library and plan and recommend paths for tracking vehicles based on the acquired real attribute data of the park; the path planning module includes a path planning unit and a path recommendation unit;

[0095] The path planning unit is used to construct a path library, and use the path optimization model built into the path library to perform secondary planning adjustments on the constructed path according to the real attribute data of the park, the smoothness of the corresponding area at different time points, and the detected vehicle types; further, the specific steps of constructing the path optimization model include:

[0096] D1. Based on the acquired park's real attribute data, the divided park areas, and the vehicle types and number of each type of vehicles in the corresponding divided areas obtained by the vehicle classification unit, a path optimization function is constructed to minimize the total time for each type of vehicle to pass through all divided areas. The specific objective function is:

[0097] ,

[0098] in, represents the decision variable for the j-th type of vehicle to use the i-th path in the n-th partition area, where If it is equal to 1, it means that the path is selected; otherwise it is 0; represents the occupied capacity of the i-th path of the j-th type of vehicle in the n-th partition area, Indicates the total number of areas divided into the park. Represents the total number of driving paths in all divided areas. Indicates the total number of vehicle types in each divided area detected by the monitoring; further, in this embodiment, the first constraint condition indicates that each type of vehicle must and can only choose one path to reach the destination at the current moment; the second constraint condition indicates that the total number of vehicles of the corresponding type on each path cannot exceed its maximum capacity; the third constraint condition indicates that the smoothness of the path selected by the decision variable must meet the minimum threshold condition of the smoothness of the path recommended by the path recommendation unit, and is the path with the largest smoothness among all existing passable paths;

[0099] D2. Build a path optimization model based on BP neural network optimization particle swarm algorithm, and build the constructed path optimization function into the particle swarm algorithm in the path optimization model as the fitness function of the particle swarm algorithm;

[0100] D3. Input the real attribute data of the park obtained in real time by video surveillance and the dynamically annotated path data in the existing path library into the path optimization model, obtain the optimized path library, and feed the optimized path library back to the path recommendation unit, so that each type of vehicle can reach the target point in the shortest time.

[0101] The path recommendation unit is used to recommend an optimal path to the currently tracked and located vehicle through the constructed path library according to the current position of the vehicle, the vehicle detection classification result, the target position and the real attribute data of the park at the current moment; further, the specific steps of the path recommendation unit recommending the optimal path in this embodiment include:

[0102] C1. Set the threshold of the smoothness coefficient and , the current smoothness coefficient of the i-th traffic path of the j-th vehicle in the n-th partition area is obtained and and For comparison, When , the j-th vehicle passes through the i-th traffic path in the n-th divided area;

[0103] C2. When When , we can calculate and Compare the travel time of the corresponding travel path with the current time smoothness coefficient and travel time corresponding to the travel paths in the remaining divided areas in the path library;

[0104] C21, when all comparison results meet and When , the j-th vehicle still travels along the i-th path in the n-th divided area until it reaches the target point, where It represents the smoothness coefficient of the j-th vehicle traveling on the v-th traffic path in the u-th divided area at the current moment, represents the real-time travel time of the j-th type of vehicle on the v-th travel path in the u-th area;

[0105] C22, when at least one of the paths in the divided area satisfies When and satisfaction The travel time corresponding to the conditional travel path When there is only one pass path that satisfies When The corresponding travel path is used as the regional travel path at the current time, and the C22 process is repeated until the target point is reached;

[0106] C23, when at least two paths satisfy When The travel paths are sorted in descending order according to the travel time, and the travel path with the smallest travel time after descending order is selected as the regional driving path at the current time. Repeat the C23 process until the target point is reached.

[0107] C24, when all the paths satisfy When , the j-th vehicle still travels through the i-th path in the n-th divided area until it reaches the target point.

[0108] The process first sets the smooth coefficient threshold so that the system can quickly determine whether the vehicle should continue to drive along the current path, avoiding unnecessary path changes and maintaining the continuity of driving. When the smooth coefficient is lower than the threshold, the system enters a more detailed path optimization stage, and provides alternative paths for vehicles by comparing the smooth coefficients and travel times of all divided areas in the park. Secondly, when all paths are unobstructed, the vehicle will continue to drive along the original route, simplifying the driver's operation and ensuring smooth driving. This series of intelligent decisions not only significantly reduces traffic congestion in the park, but also combines real-time travel time to ensure the optimality and timeliness of the recommended path.

[0109] The interactive module is used for the training and deployment of models in the vehicle positioning and tracking system and the visual display of the system analysis structure; the interactive module includes a model training and deployment unit and a visualization unit;

[0110] The model training and deployment unit is used for the training of models in each unit in the system and the one-click deployment of the training completion unit; the visualization unit is used for real-time visualization of the processing results of each unit in the system.

[0111] Example 2

[0112] See also Figure 2 , another embodiment provided by the present invention: the workflow of the corresponding unit of the smart park vehicle positioning and tracking system based on video surveillance includes:

[0113] First, the data collection unit collects the work park area map and the park's real attribute data, and inputs the collected data into the area division unit to divide the park into regions. According to the divided regions, the regional evaluation unit evaluates and predicts the smoothness of the corresponding regions at different time points. According to the real attribute data of the divided regions and the smoothness at different time points, the regional annotation unit dynamically classifies and annotates each region.

[0114] Second, based on the video data obtained from the corresponding divided areas, the multi-dimensional detection and matching model deployed in the vehicle classification unit by the model training deployment unit is used to detect and classify the vehicle, and the classified vehicle is subjected to multi-target recognition and tracking and positioning by the tracking and positioning unit;

[0115] Third, a path library is constructed through the path planning unit, and the built-in path optimization model of the path library is used to perform secondary planning adjustments on the constructed path according to the real attribute data of the park, the smoothness of the corresponding area at different time points, and the detected vehicle types. At the same time, the path recommendation unit recommends the existing or optimized travel path in the path library to the tracked and located vehicle, so that the vehicle can reach the target point in the shortest time;

[0116] Fourth, the operation results of each unit from the first to the third are displayed in real time through the visualization unit.

[0117] Example 3

[0118] See also Figure 3 Another embodiment provided by the present invention is a method for positioning and tracking vehicles in a smart park based on video surveillance, the steps comprising:

[0119] S1. Obtain the park area map and the park's real attribute data and pre-process them, and use the grid algorithm to divide the area according to the saved data. At the same time, use the evaluation algorithm to obtain the smoothness of each area at different times after the division;

[0120] S2. Dynamically classify and label each divided area according to the obtained smoothness of each area at different times;

[0121] S3. According to the video data obtained from the corresponding divided areas, the vehicles are detected and classified through the built-in multi-dimensional detection and matching model, and each type of classified vehicle is tracked and located in real time through the multi-target tracking and positioning algorithm;

[0122] S4. Construct a path library with a path optimization model, and recommend existing or optimized travel paths to the corresponding tracking vehicles through a recommendation algorithm and the path library based on the current position of the corresponding tracking vehicle, the target point position and the real attribute data of the park at the current moment, so that the vehicle can reach the target point in the shortest time.

[0123] Example 4

[0124] A computer-readable storage medium stores computer instructions, which, when executed, execute a method for positioning and tracking vehicles in a smart park based on video surveillance.

[0125] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements a smart park vehicle positioning and tracking method based on video surveillance when executing the computer program.

[0126] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

[0127] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. The smart park vehicle positioning and tracking system based on video surveillance is characterized by: include: Dynamic grid module, vehicle positioning module and path planning module; The dynamic grid module includes a data acquisition unit, a region division unit, a region evaluation unit and a region marking unit; The data collection unit is used to collect the work park area map and the park's real attribute data, and pre-process and save them; the area division unit is used to divide the area through a grid algorithm according to the acquired park area map and the park's real attribute data; The area evaluation unit is used to evaluate the smoothness of the divided corresponding area through an evaluation algorithm according to the traffic flow data at different time points in the real attribute data of the park, and obtain the smoothness of the corresponding area at different time points; the area labeling unit is used to dynamically classify and label the driving paths in the divided area according to the obtained smoothness of the corresponding area at different time points, and update the labeled paths to the path library constructed by the path planning unit; The vehicle positioning module includes a vehicle classification unit and a tracking and positioning unit; the vehicle classification unit is used to detect and classify vehicles based on the vehicle video data captured by the monitoring in real time within the divided area through the built-in multi-dimensional detection and matching model; the tracking and positioning unit is used to track and locate the classified vehicles in real time through the built-in multi-target tracking and positioning algorithm; The path planning module includes a path planning unit and a path recommendation unit; The path planning unit is used to construct a path library, and use the path optimization model built into the path library to perform secondary planning adjustments on the constructed path according to the real attribute data of the park, the smoothness of the corresponding area at different time points and the detected vehicle types; the path recommendation unit is used to recommend the optimal path to the currently tracked and located vehicle through the constructed path library according to the current position of the vehicle, the vehicle detection classification result, the target point position and the real attribute data of the park at the current moment; The specific steps of constructing the path optimization model include: D1. Based on the acquired park's real attribute data, the divided park areas, and the vehicle types and number of each type of vehicles in the corresponding divided areas obtained by the vehicle classification unit, a path optimization function is constructed to minimize the total time for each type of vehicle to pass through all divided areas. The specific objective function is: ， in, represents the decision variable for the j-th type of vehicle to use the i-th path in the n-th partition area, represents the occupied capacity of the i-th path of the j-th type of vehicle in the n-th partition area, Indicates the total number of areas divided into the park. Represents the total number of driving paths in all divided areas. Indicates the total number of vehicle types detected by the monitoring in each divided area; represents the road capacity of the ith travel path in the nth partition area, It represents the mean value of the smoothness coefficient of the i-th traffic path of the j-th vehicle type in the n-th divided area within the next 5 minutes. represents the smoothness coefficient threshold, It represents the smoothness coefficient of the j-th vehicle traveling on the v-th traffic path in the u-th divided area at the current moment, The real-time travel time of the j-th type of vehicle on the i-th travel path in the n-th area; D2. Build a path optimization model based on BP neural network optimization particle swarm algorithm, and build the constructed path optimization function into the particle swarm algorithm in the path optimization model as the fitness function of the particle swarm algorithm; D3. Input the real attribute data of the park obtained in real time by video surveillance and the dynamically annotated path data in the existing path library into the path optimization model, obtain the optimized path library, and feed the optimized path library back to the path recommendation unit, so that each type of vehicle can reach the target point in the shortest time.

2. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 1 is characterized in that: The real attribute data of the park includes: path distribution data within the park, building and impassable area data, vehicle flow video data, video surveillance distribution data, single video surveillance coverage and target point location; the specific steps of the regional assessment unit to obtain the smoothness of the corresponding area at different time points include: B1. Obtain the road capacity of the ith traffic path in the nth divided area within the three-month historical working time of the park , historical traffic volume of the jth type of vehicles , the average travel time of the j-th type of vehicle on the i-th travel path and path length , the average speed of the j-th type of vehicle when traveling smoothly on the i-th traffic path , and pre-process the acquired data; B2, dividing the pre-processed data obtained in B1 into windows of 10 minutes in length, and obtaining the divided road driving status data; B3. Using the divided road driving status data, calculate the flow density ratio corresponding to the current density and the blocking density and the driving speed ratio corresponding to the current speed and the free-flowing speed in each divided time window, specifically: , ,in, represents the travel speed ratio of the jth vehicle traveling on the i-th traffic path in the n-th divided area, represents the flow density ratio of the jth type of vehicle traveling on the i-th traffic path in the n-th divided area, represents the blocking density on the ith path in the nth partition area, represents the body width of the j-th vehicle, Represents the average width of the i-th path in the n-th partition area.

3. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 2 is characterized in that: The specific steps of the regional evaluation unit obtaining the smoothness of the corresponding region at different time points also include: B4. Based on the calculated driving speed ratio and flow density ratio, the smoothness coefficient of the j-th vehicle traveling on the i-th traffic path in the n-th divided area is calculated. , and align the calculated smooth coefficient with the driving status data collected in B1 in the time dimension to obtain a smooth coefficient prediction training set; B5. Construct a smooth traffic prediction model based on a support vector machine, input the obtained smooth traffic coefficient prediction training set into the smooth traffic prediction model for training, obtain the trained smooth traffic prediction model, and embed the obtained smooth traffic prediction model into the regional evaluation unit; B6. Obtain the real attribute data and vehicle detection classification result data in the park in real time through video surveillance, and input the acquired data into the regional evaluation unit to calculate the average value of the smooth coefficient of the i-th traffic path of the j-th vehicle type in the n-th divided area within the next 5 minutes, and use the average value as the smooth coefficient of the j-th vehicle on the i-th driving path at the current moment. ; B7. Based on the calculated smoothness coefficient of the i-th path of the j-th vehicle in the n-th divided area at the current moment, the length of the i-th path , calculate the real-time travel time of the j-th type of vehicle on the i-th travel path in the n-th area The calculated smoothness coefficient and real-time travel time are fed back to the regional annotation unit to dynamically update and annotate the corresponding regional driving path in the path library constructed by the path planning unit. Specifically, .

4. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 3 is characterized in that: The specific steps of the path recommendation unit recommending the optimal path include: C1. Set the threshold of the smoothness coefficient and , the current smoothness coefficient of the i-th traffic path of the j-th vehicle in the n-th partition area is obtained and and For comparison, When , the j-th vehicle passes through the i-th traffic path in the n-th divided area; C2. When When , we can calculate and Compare the travel time of the corresponding travel path with the current time smoothness coefficient and travel time corresponding to the travel paths in the remaining divided areas in the path library; C21, when all comparison results meet and When , the j-th vehicle still travels along the i-th path in the n-th divided area until it reaches the target point, where It represents the smoothness coefficient of the j-th vehicle traveling on the v-th traffic path in the u-th divided area at the current moment, represents the real-time travel time of the j-th type of vehicle on the v-th travel path in the u-th area; C22, when at least one of the paths in the divided area satisfies When and satisfaction The travel time corresponding to the conditional travel path When there is only one pass path that satisfies When The corresponding travel path is used as the regional travel path at the current time, and the C22 process is repeated until the target point is reached; C23, when at least two paths satisfy When The travel paths are sorted in descending order according to the travel time, and the travel path with the smallest travel time after the descending order is selected as the regional driving path at the current time, and the C23 process is repeated until the target point is reached; C24, when all the paths satisfy When , the j-th vehicle still travels along the i-th path in the n-th divided area until it reaches the target point.

5. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 4 is characterized in that: The specific step of the path recommendation unit recommending the optimal path also includes: C3. When When the path planning unit is called, the path in the path library is optimized through the path optimization model according to the current vehicle type, vehicle position, target point position and the real attribute data of the park, and the mean value of the smooth coefficient and the real-time travel time within 5 minutes corresponding to the driving path in the corresponding area after optimization are obtained. ; C4. Repeat C1 and C2 processes according to the optimized path library to recommend the corresponding regional driving path to the j-th type of vehicle identified by monitoring until it reaches the target point.

6. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 5 is characterized in that: The specific steps of the vehicle classification unit performing vehicle detection through the multi-dimensional detection matching model include: E1. Perform frame processing on the vehicle flow video data in the acquired real attribute data of the park, obtain P types of vehicle image data, and annotate the corresponding license plate number, vehicle color and vehicle attribute information data on the framed vehicle image data; E2. Build a multi-dimensional detection and matching model based on the pre-trained lightweight model, input the acquired vehicle flow video data into the lightweight VIT pre-trained sub-model in the multi-dimensional detection and matching model, extract the vehicle running status and vehicle-to-vehicle distance, and obtain the vehicle running status characteristics and vehicle-to-vehicle distance; E3. Set the vehicle distance threshold, input the labeled P types of vehicle image data into the resnet-18 pre-trained sub-model at the same time, extract the static representation corresponding to each of the P types of vehicles, and obtain the static feature set corresponding to the P types of vehicles. ; E4. Compare the obtained vehicle distance with the vehicle distance threshold. and When the inter-vehicle distance of the corresponding type of vehicles is greater than the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle can be distinguished. and When the inter-vehicle distance of the corresponding type of vehicles is less than or equal to the inter-vehicle distance threshold, and Indicates that the static characteristics of the vehicle cannot be distinguished; E5. Set vehicle similarity threshold , the obtained indistinguishable vehicle static features and Input into the similarity sub-model to calculate the similarity between the static features of the vehicle , and the calculated similarity Compared with the vehicle similarity threshold, when Less than When and The static features of the dissimilar vehicles are concatenated with the static features of distinguishable vehicles to obtain secondary distinguishable vehicle static features; E6. When Greater than or equal to When and For similar vehicle static features, the obtained similar vehicle static features are cascaded to obtain a secondary indistinguishable vehicle static feature set.

7. The smart park vehicle positioning and tracking system based on video surveillance as claimed in claim 6 is characterized in that: The specific steps of the vehicle classification unit performing vehicle detection through the multi-dimensional detection matching model also include: E7. Input the vehicle attribute information data into the Chinese pre-trained Bert sub-model to obtain the corresponding vehicle text representation features; E8. Input the acquired text representation features, the secondary indistinguishable vehicle static feature set and the vehicle running state features into the fusion attention sub-model constructed by the cross attention network to obtain the corresponding vehicle fusion distinguishing features; E9. Set the cross entropy loss function and training cycle, input the obtained vehicle fusion distinguishing features and secondary distinguishable vehicle static features into the output sub-model constructed by the softmax function for training, obtain the trained multi-dimensional detection and matching model, and output the corresponding vehicle type, vehicle width, vehicle attribute information and the number of each type of vehicle in the corresponding divided area; E10. The trained multi-dimensional detection and matching model is built into the path optimization model. The traffic flow video data in each divided area is captured in real time through video surveillance. The vehicle classification unit is used to identify and detect the vehicle type in the traffic flow video. The detected vehicle type and vehicle attribute information are fed back to the tracking and positioning unit and the path recommendation unit for positioning tracking and path recommendation.

8. A method for positioning and tracking vehicles in a smart park based on video surveillance, which is implemented based on the system for positioning and tracking vehicles in a smart park based on video surveillance according to any one of claims 1 to 7, characterized in that the steps include: S1. Obtain the park area map and the park's real attribute data and pre-process them, and use the grid algorithm to divide the area according to the saved data. At the same time, use the evaluation algorithm to obtain the smoothness of each area at different times after the division; S2. Dynamically classify and label each divided area according to the obtained smoothness of each area at different times; S3. According to the video data obtained from the corresponding divided areas, the vehicles are detected and classified through the built-in multi-dimensional detection and matching model, and each type of classified vehicle is tracked and located in real time through the multi-target tracking and positioning algorithm; S4. Construct a path library with a path optimization model, and recommend existing or optimized travel paths to the corresponding tracking vehicles through a recommendation algorithm and the path library based on the current position of the corresponding tracking vehicle, the target point position and the real attribute data of the park at the current moment.

9. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, the method for positioning and tracking vehicles in a smart park based on video surveillance as described in claim 8 is executed.

Images