Vehicle management system based on smart traffic

Abstract

The invention discloses a vehicle management system based on intelligent traffic. The vehicle management system comprises an acquisition preprocessing module, a traffic network analysis module, a query reasoning module, a track prediction module, a comprehensive monitoring module, a path optimization module, a scheduling management module, a 3D simulation module, a data management module and a report production module. The traffic network analysis module is respectively connected with the acquisition preprocessing module, the query reasoning module, the data management module, the report production module, the 3D simulation module, the comprehensive monitoring module and the trajectory prediction module, and the acquisition preprocessing module is connected with the data management module; the track prediction module is connected with the comprehensive monitoring module and the path optimization module. The comprehensive monitoring module is connected with the path optimization module and the 3D simulation module. The path optimization module is connected with the scheduling management module; and the 3D simulation module is connected with the scheduling management module. According to the invention, countermeasures can be rapidly taken for emergencies, the influence of the emergencies on traffic flow is reduced, and the event processing time is shortened.

Description

Technical Field

[0001] The present invention relates to the field of intelligent traffic management, and in particular to a vehicle management system based on intelligent traffic. Background Art

[0002] With the acceleration of global urbanization, urban traffic problems are becoming increasingly severe. Traffic congestion, frequent traffic accidents, and environmental pollution pose a major challenge to the quality of life of urban residents and the sustainable development of cities. Traditional traffic management methods are difficult to cope with the complex and changing traffic environment, and there is an urgent need to introduce advanced technical means to achieve intelligent and refined traffic management. In order to meet these challenges, the intelligent transportation system came into being. The intelligent transportation system is a system that uses information technology, communication technology, sensing technology, control technology, etc. to comprehensively perceive, effectively control, and make intelligent decisions on the transportation system. It collects, processes, and shares traffic data to provide real-time traffic information and optimization solutions, thereby improving traffic efficiency, reducing congestion and accidents, and improving environmental quality. The current intelligent transportation system still faces many challenges in data collection, processing, and analysis. For example, the sources of traffic data are diverse and scattered, and data fusion and real-time processing are difficult; the accuracy and efficiency of traffic flow prediction and path optimization algorithms need to be improved; and the performance of cross-camera matching and target tracking technology in complex scenarios still needs to be improved. Therefore, there is an urgent need for a system that integrates multiple advanced technologies to achieve more efficient traffic management and optimization.

[0003] After searching, Chinese patent No. CN202011132037.1 discloses a vehicle management system based on smart transportation. Although this invention reduces the incidence of highway congestion and traffic accidents caused by information asymmetry and can effectively improve road traffic efficiency, it is unable to take countermeasures quickly, and the probability of events affecting traffic flow increases, and the event processing time is long. In addition, the existing vehicle management system based on smart transportation cannot accurately reflect the actual road conditions, reduces the accuracy of route planning, and is prone to deviations caused by a single data source, affecting the user's travel experience. Therefore, there is an urgent need for a vehicle management system based on smart transportation. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a vehicle management system based on intelligent transportation.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] A vehicle management system based on intelligent transportation, including an acquisition preprocessing module, a traffic network analysis module, a query reasoning module, a trajectory prediction module, a comprehensive monitoring module, a path optimization module, a scheduling management module, a 3D simulation module, a data management module and a report production module;

[0007] The acquisition preprocessing module is used to clean and preprocess the acquired raw data;

[0008] The traffic network analysis module is used to analyze vehicle flow and traffic conditions;

[0009] The query reasoning module is used for information sharing and semantic query of various modules in the system;

[0010] The trajectory prediction module is used to predict future driving routes and behavior patterns based on historical trajectory data;

[0011] The integrated monitoring module is used to monitor traffic flow and vehicle status in real time, and to identify and track several targets;

[0012] The route optimization module is used to provide optimal driving route suggestions;

[0013] The dispatch management module is used to dispatch public transportation and emergency vehicles;

[0014] The 3D simulation module is used to construct a 3D model of urban traffic and perform traffic simulation and drills;

[0015] The data management module is used to store, manage and retrieve the collected raw data;

[0016] The report generation module is used to generate various traffic analysis reports.

[0017] Furthermore, the specific steps of the traffic network analysis module for analyzing vehicle flow and traffic conditions are as follows:

[0018] S101. Clean and pre-process the collected raw data, associate the road data with the node data, form a comprehensive data set containing the relationship between roads and intersections, and associate the vehicle flow data with traffic signals, event data, road data and node data and add them to the above comprehensive data set to form a comprehensive data set containing flow information and signal and event information, set each node to represent a road intersection or key position, and set each edge to represent a road between two nodes;

[0019] S102. Construct an adjacency matrix representing the graph structure, and calculate the degree matrix of the node based on the constructed adjacency matrix, define a node feature vector for each node, which includes the attribute information of the corresponding node, construct a node feature matrix based on the node feature vector, and unify the scale of each feature vector by standardizing the node feature vector;

[0020] S103. GraphSAGE is selected as the GNN model, and the sampling and aggregation functions of the GNN model are set. The historical traffic comprehensive data set is divided into a training set, a validation set, and a test set. Then, the embedding vector of each node in each layer is calculated by sampling, aggregation, combining the characteristics of each node, and updating the node embedding vector, and the predicted value of the node is output;

[0021] S104. Calculate the loss value between the predicted value and the actual value through the cross entropy loss function, calculate the gradient of each parameter of each layer of the GNN model through PyTorch, and use the SGD optimization algorithm to update the GNN model parameters. The above process is the back propagation process; the actual value is the actual observed traffic situation;

[0022] S105. Use the test set data for forward propagation, calculate the test loss and performance indicators, adjust the model hyperparameters according to the performance on the validation set, and evaluate the generalization ability of the GNN model. The hyperparameters include the weight, bias, number of layers, and number of neurons of the GNN model;

[0023] S106. Repeat the forward propagation, back propagation and parameter update steps until the GNN model reaches the preset performance on the validation set;

[0024] S107. Import the latest graph structure into the trained GNN model. The GNN model randomly samples a number of neighbor nodes from the neighbor node set of each node, and uses an aggregation function to aggregate the sampled neighbor node features to obtain aggregate features. Then, the node's own features and the aggregate features are concatenated or weighted summed. After that, the last layer of nodes is embedded and input into the linear layer or the fully connected layer, and the predicted value of the node is output.

[0025] S108. Based on the prediction results, the flow data of each node and edge are counted to obtain the flow set, and the flow changes in different time periods and different sections are compared to identify traffic patterns and rules. By comparing the flow differences of each node, the bottleneck location where the flow is concentrated is identified, and the nodes and edges that exceed the set flow threshold and congestion threshold are found. Combined with the historical accident data of the current road and the current flow prediction value, the high-risk sections where accidents may occur are identified, and the corresponding nodes and edges are marked according to each set of prediction information.

[0026] Furthermore, the key locations described in S101 include roundabouts, toll booths, traffic signal control points, entrance and exit ramps, bus stations, public transportation hubs, school vicinity, hospital vicinity, commercial centers, large shopping centers, industrial areas, logistics centers, parking lot entrances and exits, bridges, tunnel entrances, highway service areas, and emergency rescue points.

[0027] Furthermore, the trajectory prediction module predicts the future driving route and behavior pattern in the following specific steps:

[0028] S201. Extract vehicle driving trajectory data from the data processed by the acquisition preprocessing module, divide the urban roads into discrete states according to different road forms, determine the initial state of the current vehicle according to the current vehicle position, set the probability distribution of the vehicle starting state, and use historical data to calculate the transition probability between each state;

[0029] S202. predict the state distribution after several time steps, select the state distribution with the largest probability as the predicted path according to the predicted state distribution, calculate the path probability from the initial state to each predicted state, and infer the future driving path of the vehicle according to the path probability;

[0030] S203. Define different driving behavior patterns according to the state distribution, analyze the state transition sequence based on the prediction results of the future driving path, identify the vehicle's behavior pattern, and then evaluate the performance of the GNN model through various indicators such as accuracy and recall rate. According to the error analysis, adjust the state space definition and transition probability estimation method.

[0031] The state transition sequence refers to the future state predicted based on the state of the road and the target vehicle, such as the road changing from congestion to flow, and the state of the target vehicle changing from stationary to starting;

[0032] The state space is supplemented and adjusted for the missing states based on the error analysis results. The state space is specifically defined by integrating the vehicle state and the road state, such as the road state space S = {congestion, flow, ...}, and the corresponding state information obtained from the error analysis. If a state is missing in the state space, the state is added to the state space.

[0033] Furthermore, the specific steps of the integrated monitoring module to identify and track several types of targets are as follows:

[0034] S301. Convert the collected color image into a grayscale image, and perform Gaussian filtering on the processed grayscale image to obtain a preprocessed image, extract an image without moving objects from the preprocessed image, and extract features of each pixel in the image to obtain an initial background image of the corresponding road;

[0035] S302. Initialize several sets of Gaussian distribution parameters according to each pixel position of the initial background image, calculate the matching degree between each pixel value of the current frame image and each set of Gaussian distribution, and if the pixel value falls within the Gaussian distribution μ i ±2.5σ i If the range is within , it is considered a match, otherwise it is considered a mismatch, where μ i represents the center position of the i-th Gaussian distribution, σ i represents the width of the i-th Gaussian distribution;

[0036] S303. Update the mean, variance and weight of the matched Gaussian distribution, and the weight of the unmatched Gaussian distribution. If there is no matched Gaussian distribution, create a new Gaussian distribution to replace the distribution with the smallest weight to construct the corresponding background model;

[0037] S304. Calculate the matching degree between each pixel value in each subsequent frame image and each Gaussian distribution in the background model. If the probability of the pixel value in any Gaussian distribution exceeds a preset threshold, the pixel is considered to belong to the background, and a foreground mask is generated, marking the foreground target pixel, and the foreground mask is expanded, eroded, opened, and closed to remove noise and fill holes;

[0038] S305. Input the preprocessed image into the YOLO model for detection. The YOLO model extracts the preprocessed image features through the convolution layer and predicts the detection frame, category and confidence on the feature map. Then, the overlapping detection frames are removed and only the frame with the highest confidence is retained. Then, the position and size of the detection frame are adjusted according to the output of the YOLO model, and the objects in each group of detection frames are classified to distinguish different types of vehicles and pedestrians, and a category label is assigned to each object.

[0039] S306. Predict the next position of the target through the Kalman filter, calculate the matching cost between the two targets and the target selected by the detection box, and find the optimal match through the Hungarian algorithm. Update the trajectory information of each target according to the matching result, and then extract the appearance feature vector and spatiotemporal features of the target. According to the calibration parameters and time synchronization information of the camera, align the spatiotemporal information under different cameras, calculate the Euclidean distance of the appearance feature vector and spatiotemporal features, and measure the appearance similarity and spatiotemporal consistency of the two targets;

[0040] S307. If the similarity exceeds the preset threshold, the two targets are considered to be the same target, and the trajectories of the successfully matched targets under different cameras are merged to form a complete trajectory, and the trajectory information is updated, including time, location and appearance features. The number of targets passing through the road is counted, the traffic flow is calculated, and the current traffic status is evaluated based on the calculated traffic flow and target trajectory.

[0041] Furthermore, the specific steps of the path optimization module providing optimal driving path suggestions are as follows:

[0042] S401. Collect current traffic conditions and historical traffic data, take the current state of the vehicle, including the current position and time, as the root node, and generate several groups of vehicle future states based on the prediction results of the trajectory prediction module and the current road traffic status, and connect them to the root node as child nodes, and then assign attention weights to each path according to the current traffic conditions;

[0043] S402. Starting from the root node, a child node is selected through the UCB selection strategy, and the selected path is continued to be selected downward until a leaf node is reached or a predetermined depth is reached. At each selected node, the process of the vehicle driving to the next node is simulated, and the driving time and congestion cost from the current node to the next node are calculated. During the simulation, the cost of each path is accumulated to obtain the total driving cost. The leaf node is a node that cannot be selected later.

[0044] S403. When the preset simulation time is reached, the simulation is stopped, and the total path cost from the root node to the leaf node is calculated. According to the total path cost, the quality of the path is evaluated, and then the evaluation result is back-propagated from the leaf node to the root node, and the number of visits and cumulative cost of each node along the way are updated;

[0045] S404. Repeat the selection, expansion, simulation and backtracking steps until the preset iteration time is reached, and perform weighted summation on the generated groups of paths to obtain a comprehensive evaluation. Select the path with the highest attention weight as the recommended path, and generate a specific driving path based on the results of the optimization calculation. At the same time, mark the recommended path on the map to provide driving guidance.

[0046] Furthermore, the traffic network analysis module is connected with the collection preprocessing module, the query reasoning module, the data management module, the report production module, the 3D simulation module, the comprehensive monitoring module and the trajectory prediction module.

[0047] The acquisition preprocessing module is connected with the data management module;

[0048] The trajectory prediction module is connected to the comprehensive monitoring module and the path optimization module respectively;

[0049] The comprehensive monitoring module is connected to the path optimization module and the 3D simulation module respectively;

[0050] The path optimization module is connected with the scheduling management module;

[0051] The 3D simulation module is connected with the scheduling management module.

[0052] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0053] 1. Improve event response efficiency: The present invention extracts the first N frames of images without moving objects from the video image, combines the Gaussian distribution model to construct the initial background image, and uses the YOLO model to detect and track the target. Through the spatiotemporal consistency matching of multi-camera information and the merging of target trajectories, the system can quickly identify emergencies in traffic, such as traffic accidents or road obstacles. Based on the calculation of real-time traffic flow and target trajectory, the system can quickly take countermeasures to reduce the impact of events on traffic flow and significantly shorten the event processing time, thereby ensuring that traffic quickly resumes normal operation and reduces congestion and accidents.

[0054] 2. Accurate traffic data and traffic status assessment: The system provides more accurate traffic flow data through efficient image processing and target tracking technology, and can classify and track different types of traffic targets (such as vehicles, pedestrians, etc.). Through big data analysis, it can monitor traffic status in real time, assess road traffic, congestion and potential traffic bottlenecks, thereby supporting more refined traffic management decisions and improving overall traffic efficiency.

[0055] 3. Optimize route planning and improve user experience: The present invention integrates current traffic conditions and historical data, combined with the analysis results of the trajectory prediction module, to generate multiple possible driving routes, and assigns weights to each path based on factors such as traffic flow, congestion, and driving time. The UCB selection strategy is used to optimize route planning and provide optimal driving route recommendations. By simulating the driving costs of different paths, the system can effectively avoid path planning deviations caused by a single data source, ensuring that the recommended path is more in line with actual traffic conditions, thereby improving the accuracy and convenience of travel and enhancing the user's overall travel experience.

[0056] 4. Efficient traffic flow and path optimization processing: This system uses the GNN (graph neural network) model to conduct in-depth analysis and prediction of traffic data, which can accurately calculate the flow changes of each node and identify traffic patterns. By setting flow and congestion thresholds, it can identify high-risk areas and reduce the occurrence of traffic accidents. At the same time, the system can also optimize the path according to the prediction results, reduce traffic congestion, ensure smooth traffic, and improve road traffic efficiency.

[0057] 5. Comprehensive data management and report generation: The system uses a powerful data management module to store, retrieve and manage data, and automatically generates traffic analysis reports through a report production module, so that decision makers can understand the traffic conditions and system operation effects in real time. The report content includes traffic statistics, accident analysis, route optimization effect evaluation, etc., to support urban traffic managers to make more scientific and reasonable decisions.

[0058] In summary, the present invention not only improves the efficiency and accuracy of the traffic management system, but also enhances the resilience and adaptability of the intelligent transportation system in practical applications by integrating advanced video surveillance technology, deep learning algorithms, traffic flow analysis and path optimization algorithms. It has strong real-time response and decision-making support capabilities, can effectively alleviate traffic congestion, reduce the incidence of traffic accidents, optimize road resource allocation, and greatly improve the comprehensive efficiency of urban traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a system block diagram of a vehicle management system based on intelligent transportation proposed by the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] Reference Figure 1 , this embodiment provides a vehicle management system based on smart transportation, including an acquisition preprocessing module, a traffic network analysis module, a query reasoning module, a trajectory prediction module, a comprehensive monitoring module, a path optimization module, a scheduling management module, a 3D simulation module, a data management module and a report production module.

[0062] The acquisition preprocessing module is used to clean and preprocess the collected raw data; specifically:

[0063] The acquisition preprocessing module cleans and preprocesses the collected raw data, then associates the road data with the node data to form a comprehensive data set containing the relationship between roads and intersections, and associates the vehicle flow data and traffic signals, event data, road data, and node data to the above comprehensive data set to form a comprehensive data set containing flow information and signal and event information. Each node is set to represent a road intersection or key location, and each edge is set to represent the road between two nodes.

[0064] Construct an adjacency matrix representing the graph structure, and calculate the degree matrix of the node based on the constructed adjacency matrix, define a feature vector for each node, which contains the attribute information of the node, and construct a node feature matrix based on the node feature vector. By standardizing the node feature vector, unify the feature scales;

[0065] Select GraphSAGE as the GNN model, set the sampling and aggregation functions of the GNN model, divide the historical traffic comprehensive data set into training set, validation set and test set, then calculate the embedding vector of each node in each layer through sampling, aggregation, combining the characteristics of each node and updating the node embedding vector, and output the predicted value of the node;

[0066] The loss value between the predicted value and the actual value is calculated through the cross entropy loss function. The gradient of each parameter of each layer of the GNN model is calculated through PyTorch. The GNN model parameters are updated using the SGD optimization algorithm. The above process is the back propagation process.

[0067] Use the validation set data for forward propagation, calculate the validation loss and performance indicators, and evaluate the generalization ability of the model; adjust the model hyperparameters according to the performance on the validation set. The hyperparameters include the weight, bias, number of layers, and number of neurons of the GNN model;

[0068] Repeat the forward propagation, backpropagation and parameter update steps until the model reaches the preset performance on the validation set;

[0069] Import the latest graph structure into the trained GNN model. The GNN model randomly samples multiple neighbor nodes from the neighbor node set of each node, aggregates the sampled neighbor node features using an aggregation function, and then concatenates or weighted sums the node's own features with the aggregated features. After that, the last layer of nodes is embedded and input into the linear layer or the fully connected layer, and the predicted value of the node is output.

[0070] Based on the prediction results, the flow data of each node and edge are counted to obtain the flow set, and the flow changes in different time periods and different sections are compared to identify traffic patterns and rules. By comparing the flow differences of each node, the bottleneck locations where the flow is concentrated are identified, and the nodes and edges with high flow and severe congestion are found. Combined with historical accident data and current flow prediction values, high-risk sections where accidents may occur are identified, and the corresponding nodes and edges are marked according to each set of prediction information.

[0071] It should be further explained that key locations include roundabouts, toll booths, traffic signal control points, entrance and exit ramps, bus stops, public transportation hubs, school vicinity, hospital vicinity, commercial centers, large shopping centers, industrial areas, logistics centers, parking lot entrances and exits, bridges, tunnel entrances, highway service areas and emergency rescue points.

[0072] The traffic network analysis module is used to analyze vehicle flow and traffic conditions.

[0073] The query reasoning module is used for information sharing and semantic query among modules in the system;

[0074] The trajectory prediction module is used to predict future driving routes and behavior patterns based on historical trajectory data. Specifically:

[0075] The vehicle driving trajectory data is extracted from the data processed by the acquisition preprocessing module, and the urban roads are divided into discrete states according to different road forms. According to the current position of the vehicle, its initial state is determined, the probability distribution of the vehicle's starting state is set, and the transition probability between each state is calculated using historical data; the state distribution after multiple time steps is predicted, and the state distribution with the largest probability is selected as the predicted path according to the predicted state distribution, and the path probability from the initial state to each predicted state is calculated, and the future driving path of the vehicle is inferred according to the probability of each path; different driving behavior modes are defined according to the state distribution, and the state transition sequence is analyzed based on the prediction results of the future driving path, and the vehicle's behavior mode is identified. Then, the performance of the GNN model is evaluated through various indicators such as accuracy and recall rate, and the state space definition and transition probability estimation method are adjusted according to the error analysis.

[0076] The integrated monitoring module is used to monitor traffic flow and vehicle status in real time, identify and track multiple targets, specifically:

[0077] The color image collected from the video is converted into a grayscale image, and the processed gray image is Gaussian filtered, the first N frames of images taken without moving objects are extracted from the video, and the features of each pixel in the image are extracted to obtain the initial background image of the corresponding road;

[0078] According to the position of each pixel in the initial background image, multiple sets of Gaussian distribution parameters are initialized, and the matching degree between each pixel value of the current frame image and each set of Gaussian distribution is calculated. If the pixel value falls within the Gaussian distribution μ i ±2.5σ i If the range is within , it is considered a match, otherwise it is considered a mismatch, where μ i represents the center position of the i-th Gaussian distribution, σ i represents the width of the i-th Gaussian distribution;

[0079] Update the mean, variance and weight of the matched Gaussian distribution, as well as the weight of the unmatched Gaussian distribution. If there is no matched Gaussian distribution, create a new Gaussian distribution to replace the distribution with the smallest weight to construct the corresponding background model.

[0080] Calculate the matching degree between each pixel value in each subsequent frame and each Gaussian distribution in the background model. If the probability of the pixel value in any Gaussian distribution exceeds the preset threshold, the pixel is considered to belong to the background, and a foreground mask is generated to mark the foreground target pixel. The foreground mask is expanded, eroded, opened, and closed to remove noise and fill holes.

[0081] The preprocessed image is input into the YOLO model for detection. The YOLO model extracts the preprocessed image features through the convolution layer and predicts the detection frame, category and confidence on the feature map. Then, the overlapping detection frames are removed and only the frame with the highest confidence is retained. The position and size of the detection frame are adjusted according to the output of the YOLO model. The objects in each group of detection frames are classified to distinguish different types of vehicles and pedestrians, and a category label is assigned to each object.

[0082] S306. Predict the next position of the target through the Kalman filter, calculate the matching cost between the two targets and the target selected by the detection box, and find the optimal match through the Hungarian algorithm. Update the trajectory information of each target according to the matching result, and then extract the appearance feature vector and spatiotemporal features of the target. According to the calibration parameters and time synchronization information of the camera, align the spatiotemporal information under different cameras, calculate the Euclidean distance of the appearance feature vector and spatiotemporal features, and measure the appearance similarity and spatiotemporal consistency of the two targets;

[0083] If the similarity exceeds the preset threshold, the two targets are considered to be the same target, and the trajectories of the successfully matched targets under different cameras are merged to form a complete trajectory, and the trajectory information is updated, including time, location and appearance features, and the number of targets passing through the road is counted, and the traffic flow is calculated. Based on the calculated traffic flow and target trajectory, the current traffic status is evaluated. It can quickly take countermeasures for emergencies, reduce the impact of events on traffic flow, shorten event processing time, and ensure rapid restoration of normal traffic operations. It can provide more detailed and accurate traffic flow data and support more refined traffic analysis and management.

[0084] The route optimization module is used to provide the best driving route suggestions, specifically:

[0085] Collect current traffic conditions and historical traffic data, take the current state of the vehicle, including the current position and time, as the root node, and generate multiple sets of vehicle future states based on the prediction results of the trajectory prediction module and the current road traffic status, connect them to the root node as child nodes, and then assign attention weights to each path based on the current traffic conditions;

[0086] Starting from the root node, a child node is selected through the UCB selection strategy, and the selected path is continued to be selected downward until a leaf node is reached or a predetermined depth is reached. At each selected node, the process of the vehicle driving to the next node is simulated, and the travel time and congestion cost from the current node to the next node are calculated. During the simulation, the cost of each path is accumulated to obtain the total travel cost.

[0087] When the preset simulation time is reached, the simulation is stopped and the total path cost from the root node to the leaf node is calculated. Based on the total path cost, the quality of the path is evaluated, and then the evaluation result is back-propagated from the leaf node to the root node, and the number of visits and cumulative cost of each node along the way are updated;

[0088] Repeat the selection, expansion, simulation and backtracking steps until the preset iteration time is reached, and perform weighted summation on the generated groups of paths to obtain a comprehensive evaluation, select the path with the highest attention weight as the recommended path, and generate a specific driving path based on the results of the optimization calculation. At the same time, mark the recommended path on the map to provide driving guidance. It can more accurately reflect the actual road conditions, improve the accuracy of path planning, reduce the deviation caused by a single data source, and enhance the overall travel experience of users.

[0089] The dispatch management module is used to intelligently dispatch public transportation and emergency vehicles;

[0090] The 3D simulation module is used to build a 3D model of urban traffic and conduct traffic simulation and drills;

[0091] The data management module is used to store, manage and retrieve the collected data;

[0092] The report generation module is used to generate various traffic analysis reports.

[0093] The present invention is not limited to the embodiments described above. The above description of the specific embodiments is intended to describe and illustrate the technical solution of the present invention. The above specific embodiments are merely illustrative and not restrictive. Without departing from the scope of the present invention and the scope of protection of the claims, a person of ordinary skill in the art can also make many forms of specific changes under the guidance of the present invention, which all fall within the scope of protection of the present invention.

Claims

1. A vehicle management system based on intelligent transportation, characterized in that: It includes collection preprocessing module, traffic network analysis module, query reasoning module, trajectory prediction module, comprehensive monitoring module, path optimization module, scheduling management module, 3D simulation module, data management module and report production module; The acquisition preprocessing module is used to clean and preprocess the acquired raw data; The traffic network analysis module is used to analyze vehicle flow and traffic conditions; The query reasoning module is used for information sharing and semantic query of various modules in the system; The trajectory prediction module is used to predict future driving routes and behavior patterns based on historical trajectory data; The integrated monitoring module is used to monitor traffic flow and vehicle status in real time, and to identify and track several targets; The route optimization module is used to provide optimal driving route suggestions; The dispatch management module is used to dispatch public transportation and emergency vehicles; The 3D simulation module is used to construct a 3D model of urban traffic and perform traffic simulation and drills; The data management module is used to store, manage and retrieve the collected raw data; The report generation module is used to generate various traffic analysis reports.

2. A vehicle management system based on intelligent transportation according to claim 1, characterized in that: The specific steps of analyzing vehicle flow and traffic conditions by the traffic network analysis module are as follows: S101. Clean and pre-process the collected raw data, associate the road data with the node data, form a comprehensive data set containing the relationship between roads and intersections, and associate the vehicle flow data with traffic signals, event data, road data and node data and add them to the above comprehensive data set to form a comprehensive data set containing flow information and signal and event information, set each node to represent a road intersection or key position, and set each edge to represent a road between two nodes; S102. Construct an adjacency matrix representing the graph structure, and calculate the degree matrix of the node based on the constructed adjacency matrix, define a node feature vector for each node, which includes the attribute information of the corresponding node, construct a node feature matrix based on the node feature vector, and unify the scale of each feature vector by standardizing the node feature vector; S103. GraphSAGE is selected as the GNN model, and the sampling and aggregation functions of the GNN model are set. The historical traffic comprehensive data set is divided into a training set, a validation set, and a test set. Then, the embedding vector of each node in each layer is calculated by sampling, aggregation, combining the characteristics of each node, and updating the node embedding vector, and the predicted value of the node is output; S104. Calculate the loss value between the predicted value and the actual value through the cross entropy loss function, calculate the gradient of each parameter of each layer of the GNN model through PyTorch, and use the SGD optimization algorithm to update the GNN model parameters. The above process is the back propagation process; The actual value is the actual observed traffic situation; S105. Use the test set data for forward propagation, calculate the test loss and performance indicators, adjust the model hyperparameters according to the performance on the validation set, and evaluate the generalization ability of the GNN model. The hyperparameters include the weight, bias, number of layers, and number of neurons of the GNN model; S106. Repeat the forward propagation, back propagation and parameter update steps until the GNN model reaches the preset performance on the validation set; S107. Import the latest graph structure into the trained GNN model. The GNN model randomly samples a number of neighbor nodes from the neighbor node set of each node, and uses an aggregation function to aggregate the sampled neighbor node features to obtain aggregate features. Then, the node's own features and the aggregate features are concatenated or weighted summed. After that, the last layer of nodes is embedded and input into the linear layer or the fully connected layer, and the predicted value of the node is output. S108. Based on the prediction results, the flow data of each node and edge are counted to obtain the flow set, and the flow changes in different time periods and different sections are compared to identify traffic patterns and rules. By comparing the flow differences of each node, the bottleneck location where the flow is concentrated is identified, and the nodes and edges that exceed the set flow threshold and congestion threshold are found. Combined with the historical accident data of the current road and the current flow prediction value, the high-risk sections where accidents may occur are identified, and the corresponding nodes and edges are marked according to each set of prediction information.

3. A vehicle management system based on intelligent transportation according to claim 2, characterized in that: The key locations described in S101 include roundabouts, toll booths, traffic signal control points, entrance and exit ramps, bus stations, public transportation hubs, school vicinity, hospital vicinity, commercial centers, large shopping malls, industrial areas, logistics centers, parking lot entrances and exits, bridges, tunnel entrances, highway service areas, and emergency rescue points.

4. A vehicle management system based on intelligent transportation according to claim 1, characterized in that: The specific steps of the trajectory prediction module predicting the future driving route and behavior pattern are as follows: S201. Extract vehicle driving trajectory data from the data processed by the acquisition preprocessing module, divide the urban roads into discrete states according to different road forms, determine the initial state of the current vehicle according to the current vehicle position, set the probability distribution of the vehicle starting state, and use historical data to calculate the transition probability between each state; S202. predict the state distribution after several time steps, select the state distribution with the largest probability as the predicted path according to the predicted state distribution, calculate the path probability from the initial state to each predicted state, and infer the future driving path of the vehicle according to the path probability; S203. Define different driving behavior patterns according to the state distribution, analyze the state transition sequence according to the prediction results of the future driving path, identify the vehicle's behavior pattern, and then evaluate the performance of the GNN model through various indicators such as accuracy and recall rate, and adjust the state space definition and transition probability estimation method according to the error analysis; The state transition sequence refers to the future state predicted based on the state of the road and the target vehicle; the state space is to supplement and adjust the missing state according to the error analysis results, by integrating the definitions of vehicle state and road state, and the error analysis results correspond State information, if missing in the state space A state , then add the state to the state space.

5. The vehicle management system based on intelligent transportation according to claim 1 is characterized in that: The specific steps of the integrated monitoring module to identify and track several types of targets are as follows: S301. Convert the collected color image into a grayscale image, and perform Gaussian filtering on the processed grayscale image to obtain a preprocessed image, extract an image without moving objects from the preprocessed image, and extract features of each pixel in the image to obtain an initial background image of the corresponding road; S302. Initialize several sets of Gaussian distribution parameters according to each pixel position of the initial background image, calculate the matching degree between each pixel value of the current frame image and each set of Gaussian distribution, and if the pixel value falls within the Gaussian distribution μ i ±2.5σ i If the range is within , it is considered a match, otherwise it is considered a mismatch, where μ i represents the center position of the i-th Gaussian distribution, σ i represents the width of the i-th Gaussian distribution; S303. Update the mean, variance and weight of the matched Gaussian distribution, and the weight of the unmatched Gaussian distribution. If there is no matched Gaussian distribution, create a new Gaussian distribution to replace the distribution with the smallest weight to construct the corresponding background model; S304. Calculate the matching degree between each pixel value in each subsequent frame image and each Gaussian distribution in the background model. If the probability of the pixel value in any Gaussian distribution exceeds a preset threshold, the pixel is considered to belong to the background, and a foreground mask is generated, marking the foreground target pixel, and the foreground mask is expanded, eroded, opened, and closed to remove noise and fill holes; S305. Input the preprocessed image into the YOLO model for detection. The YOLO model extracts the preprocessed image features through the convolution layer and predicts the detection frame, category and confidence on the feature map. Then, the overlapping detection frames are removed and only the detection frame with the highest confidence is retained. Then, the position and size of the detection frame are adjusted according to the output of the YOLO model, and the objects in each group of detection frames are classified to distinguish different types of vehicles and pedestrians, and a category label is assigned to each object. S306. Predict the next position of the target through the Kalman filter, calculate the matching cost between the two targets and the target selected by the detection box, and find the optimal match through the Hungarian algorithm. Update the trajectory information of each target according to the matching result, and then extract the appearance feature vector and spatiotemporal features of the target. According to the calibration parameters and time synchronization information of the camera, align the spatiotemporal information under different cameras, calculate the Euclidean distance of the appearance feature vector and spatiotemporal features, and measure the appearance similarity and spatiotemporal consistency of the two targets; S307. If the similarity exceeds the preset threshold, the two targets are considered to be the same target, and the trajectories of the successfully matched targets under different cameras are merged to form a complete trajectory, and the trajectory information is updated, including time, location and appearance features. The number of targets passing through the road is counted, the traffic flow is calculated, and the current traffic status is evaluated based on the calculated traffic flow and target trajectory.

6. A vehicle management system based on intelligent transportation according to claim 1, characterized in that: The specific steps of the path optimization module providing the optimal driving path suggestion are as follows: S401. Collect current traffic conditions and historical traffic data, take the current state of the vehicle, including the current position and time, as the root node, and generate several groups of vehicle future states based on the prediction results of the trajectory prediction module and the current road traffic status, and connect them to the root node as child nodes, and then assign attention weights to each path according to the current traffic conditions; S402. Starting from the root node, a child node is selected through the UCB selection strategy, and the selected path is continued to be selected downward until a leaf node is reached or a predetermined depth is reached. At each selected node, the process of the vehicle driving to the next node is simulated, and the driving time and congestion cost from the current node to the next node are calculated. During the simulation process, the cost of each path segment is accumulated to obtain the total driving cost; Leaf nodes are nodes that cannot be subsequently selected; S403. When the preset simulation time is reached, the simulation is stopped, and the total path cost from the root node to the leaf node is calculated. According to the total path cost, the quality of the path is evaluated, and then the evaluation result is back-propagated from the leaf node to the root node, and the number of visits and cumulative cost of each node along the way are updated; S404. Repeat the selection, expansion, simulation and backtracking steps until the preset iteration time is reached, and perform weighted summation on the generated groups of paths to obtain a comprehensive evaluation. Select the path with the highest attention weight as the recommended path, and generate a specific driving path based on the results of the optimization calculation. At the same time, mark the recommended path on the map to provide driving guidance.

7. A vehicle management system based on intelligent transportation according to claim 1, characterized in that: The traffic network analysis module is connected with the collection preprocessing module, query reasoning module, data management module, report production module, 3D simulation module, comprehensive monitoring module and trajectory prediction module respectively. The acquisition preprocessing module is connected with the data management module; The trajectory prediction module is connected to the comprehensive monitoring module and the path optimization module respectively; The comprehensive monitoring module is connected to the path optimization module and the 3D simulation module respectively; The path optimization module is connected with the scheduling management module; The 3D simulation module is connected with the scheduling management module.

Images