Road traffic risk management and control system and method based on data analysis

Abstract

This invention relates to the field of road management, specifically to a data-driven road traffic risk control system and method, comprising: a road monitoring module, a vehicle management module, an accident analysis module, a traffic focusing module, and a dispatching and allocation module. The road monitoring module is used to establish a road network model; the vehicle management module is used to plan the optimal driving route; the accident analysis module is used to determine the type of accident; the traffic focusing module is used to optimize patrol routes; and the dispatching and allocation module is used to build an exchange service platform for command and dispatch. This invention can improve traffic operation efficiency, reduce accident risks, build an intelligent traffic management system, enhance the perception and management level of urban traffic, improve the accuracy of traffic accident target detection, improve traffic infrastructure construction, enhance road network safety, optimize the allocation of human and material resources, and reduce traffic management costs.

Description

Technical Field

[0001] This invention relates to the field of road management, specifically to a road traffic risk management system and method based on data analysis. Background Technology

[0002] A road traffic management system is a system implemented by traffic management departments to organize and control vehicles and the traffic environment on roads in order to ensure traffic order, safety, and smooth traffic flow. It also provides functions such as accident tracing and road guidance services. Common road management methods involve statistical processing of traffic data to analyze the impact of vehicle, road condition, and environmental factors on traffic, thereby achieving the goal of traffic command and dispatch.

[0003] Due to the dynamic and random nature of real-time road conditions, the accuracy and completeness of data vary, leading to high complexity in road traffic decision-making. As urban traffic conditions become increasingly complex, complex scene changes will affect the performance of road cameras. Existing monitoring systems lack detailed modeling of road networks, have low access to real-time information, slow recognition processes, and lagging processing, thus failing to provide real-time guidance for traffic management.

[0004] Furthermore, traffic accidents are subject to complex correlations, and traffic accident analysis suffers from problems such as low solution efficiency, insufficient long-term stability and generalization ability. This results in insufficient accuracy and effectiveness in commanding and dispatching limited management resources, making it impossible to achieve high dispatch efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a road traffic risk management system and method based on data analysis to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: a road traffic risk management system based on data analysis, comprising: a road monitoring module, a vehicle management module, an accident analysis module, a traffic focusing module, and a dispatching and allocation module; The road monitoring module is used to collect road monitoring images, identify all vehicles, obtain road speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. The vehicle management module is used to establish a path planning architecture on the road network model, select paths based on road network traffic conditions and vehicle trajectories, set road segment rewards based on traffic density of each road segment, plan the optimal driving path through vehicle travel time and trajectory rewards, and distribute the results to each vehicle. The accident analysis module is used to extract accident features from traffic police data, construct a road traffic accident dataset, output a prediction frame through a latent diffusion model based on the optical flow frame and video frame of the monitoring image, compare the pixel gray level difference between the prediction frame and the current frame, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster the principal component scores to obtain the average cluster comprehensive score, and determine the occurrence and type of the accident. The traffic focusing module is used to extract the spatiotemporal features of hotspot areas from road environment data to establish spatial connection relationships, identify hotspot areas based on hotspot analysis, calculate the spatiotemporal kernel density value of traffic accidents, analyze the distribution characteristics of traffic violations from a spatiotemporal dimension, and optimize patrol routes based on distribution characteristics to cover traffic hotspot areas. The scheduling and allocation module is used to connect media terminals of different standards to the scheduling platform, build an exchange service platform for command and dispatch, construct a two-dimensional directed network model, quantify the optimization degree of each facility and service for road accidents, construct an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

[0007] Furthermore, the road monitoring module includes: a spatial conversion unit, an information access unit, and a network modeling unit; The spatial transformation unit is used to solve the camera's intrinsic parameters, including focal length, pitch angle, and yaw angle, through perspective geometric constraints. It establishes a self-calibration model based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and the actual position, and initializes the model parameters using road markings, vehicle trajectories, and road scenes. The information access unit is used to set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the operating conditions, divide the optimization range of the parameters according to the operating conditions, obtain real-time traffic flow data, use a distributed message queue to cache the real-time traffic flow, and build a dynamic information database for urban traffic. The network modeling unit is used to convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

[0008] Furthermore, the vehicle management module includes: a data association unit, a trajectory analysis unit, and a road segment planning unit; The data association unit is used to construct a database with traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the least support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. The trajectory analysis unit is used to process vehicle trajectories using a segmented embedding method, converting the trajectory embedding into a text feature representation, inputting it into an LLM for prediction, and embedding task requirements, map semantics, and environmental information into the model to determine the environmental input. The road segment planning unit uses environmental input as state, trajectory change as action, and the difference in average vehicle waiting time between time steps as reward. It calculates the reward function, outputs real-time road segment reward, and gives road segment reward related to the time spent to each node reached by the vehicle. The optimal path is calculated with the goal of maximizing reward and then distributed to the vehicle.

[0009] Furthermore, the accident analysis module includes: an image comparison unit and an accident classification unit; The image comparison unit is used to extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. It diffuses the optical flow in potential spaces of two different scales, fuses motion information and appearance information, and fuses and decodes the output to generate prediction frames. It calculates the pixel grayscale difference between the prediction frames and the real frames to obtain a difference map, performs principal component analysis on the difference map, and obtains the principal component load matrix. The accident classification unit is used to project the difference map onto the principal component space to obtain the principal component score vector. The variance contribution rate of each principal component is calculated as a weight to weight the score, resulting in a weighted score vector. The principal component score vectors of historical accident samples are used to train a clustering model. The principal component score vector of the current abnormal event is compared with the mean of each cluster to calculate the similarity and determine the accident type.

[0010] Furthermore, the traffic focusing module includes: a heat map connection unit and a feature optimization unit; The heat connection unit is used to establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, the local sum of the accident frequency with neighboring units is calculated and compared with the global expectation to calculate the hot spot and cold spot classification of the accident at different locations. The feature optimization unit is used to convert discrete hotspot density into a continuous probability density surface, generate a spatial kernel density map, identify spatiotemporally continuous clusters using spatiotemporal scanning statistics, cluster shape features and density features are improved by the DBSCAN algorithm, fine-grained division of cluster boundaries is performed to obtain road segment weights, and the maximum coverage path algorithm is used to plan patrol routes.

[0011] Furthermore, the scheduling and allocation module includes: a command platform unit and an element scheduling unit; The command platform unit is used to connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. The element scheduling unit is used to align multi-source data in time and space, classify accident risk levels, calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence through the weighted HITS algorithm as quantitative indicators of operational performance, and use a binomial Logit model to fit the influence weights of each influencing factor on the probability of traffic violation events.

[0012] The data-driven approach to road traffic risk management includes the following steps: Step S1. Collect road monitoring images, identify all vehicles, obtain road travel speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. Step S2. Establish a path planning architecture on the road network model, set road segment rewards according to the traffic density of each road segment, plan the optimal driving route through vehicle travel time and trajectory rewards, and distribute it to each vehicle; Step S3. Extract accident features from traffic police data, construct a road traffic accident dataset, calculate the pixel grayscale difference between the predicted frame and the current frame of the monitoring image, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster to obtain the average cluster comprehensive score, and determine the accident type. Step S4. Identify hotspot areas based on hotspot analysis, establish spatial connectivity, calculate the spatiotemporal kernel density value of traffic accidents, and optimize patrol routes based on the distribution characteristics of traffic violations to cover traffic hotspot areas; Step S5. Connect different media terminals to the dispatch platform, build an exchange service platform for command and dispatch, quantify the optimization degree of each facility and business for road accidents, build an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

[0013] Furthermore, step S1 includes: Step S11. Solve the camera intrinsic parameters, including focal length, pitch angle and yaw angle, through perspective geometric constraints. Based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and actual position, establish a self-calibration model and initialize the model parameters using road markings, vehicle trajectories and road scenes. Step S12. Set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the working conditions, divide the optimization interval of the parameters according to the working conditions, obtain real-time traffic flow data, use the distributed message queue to cache the real-time traffic flow, and build a dynamic information database of urban traffic. Step S13. Convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

[0014] Furthermore, step S2 includes: Step S21. Construct a database using traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the lowest support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. Step S22. Process the vehicle trajectory using a segmented embedding method, convert the trajectory embedding into a text feature representation, input it into LLM for prediction, embed the task requirements, map semantics and environmental information into the model, determine the environmental input, the environmental input is the state, the trajectory change is the action, the difference in the average waiting time of the vehicle before and after is the reward, calculate the reward function, output the real-time road segment reward, give the vehicle a road segment reward related to the time spent when it reaches a node, calculate the optimal path with the goal of maximizing the reward, and send it to the vehicle.

[0015] Furthermore, step S3 includes: Step S31. Extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. Diffusion is performed in potential spaces of two different scales to fuse motion information and appearance information. The output is then fused and decoded to generate prediction frames. The pixel grayscale difference between the prediction frames and the real frames is calculated to obtain a difference map. Principal component analysis is performed on the difference map to obtain the principal component load matrix. Step S32. Project the difference map onto the principal component space to obtain the principal component score vector. Calculate the variance contribution rate of each principal component as a weight to weight the score and obtain a weighted score vector. Use the principal component score vectors of historical accident samples to train a clustering model. Compare the principal component score vector of the current abnormal event with the mean of each cluster to calculate the similarity and determine the accident type.

[0016] Furthermore, step S4 includes: Step S41. Establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, calculate the local sum of the accident frequencies with neighboring units and compare it with the global expectation. Calculate the hot spot and cold spot classification of the accident at different locations. Step S42. Convert discrete hotspot density into continuous probability density surface, generate spatial kernel density map, identify spatiotemporally continuous clusters using spatiotemporal scanning statistics, refine the cluster shape and density features by improving the DBSCAN algorithm, perform fine-grained division of cluster boundaries, obtain road segment weights, and use the maximum coverage path algorithm to plan patrol routes.

[0017] Furthermore, step S5 includes: Step S51. Connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. Step S52. Spatiotemporally align multi-source data, classify accident risk levels, and calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence using the weighted HITS algorithm as quantitative indicators of operational performance, and fit the influence weights of each influencing factor on the probability of traffic violation events using a binomial Logit model.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are: 1. This invention establishes a road network model by constructing the spatial coordinate relationship of pixels in monitoring equipment, setting road segment rewards based on road network traffic conditions, planning the optimal driving path, updating the neural operator network based on real-time traffic flow data, and adjusting the path planning strategy. This can improve traffic operation efficiency, reduce accident risks, build an intelligent traffic management system, and enhance the overall perception and management level of urban traffic.

[0019] 2. This invention extracts accident features from traffic police data, calculates the scores of each accident principal component, determines the occurrence and type of the accident, and can realize the function of identifying traffic accidents and dynamic planning, improve the accuracy of traffic accident target detection, enhance the ability to evaluate traffic accident risks in time and space dimensions, and improve the level and governance capacity of road traffic safety management.

[0020] 3. This invention identifies hotspot areas from road environment data, optimizes patrol routes based on the distribution characteristics of traffic violations, and conducts command and dispatch according to the intensity level of identified elements. This can maintain high dispatch efficiency, improve traffic infrastructure construction, enhance road network safety, optimize the allocation of human and material resources, and reduce traffic management costs. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a schematic diagram of the road traffic risk management system based on data analysis according to the present invention; Figure 2 This is a schematic diagram illustrating the steps of the road traffic risk management method based on data analysis of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1 to 2 The present invention provides a technical solution: a road traffic risk management and control system based on data analysis, comprising: a road monitoring module, a vehicle management module, an accident analysis module, a traffic focusing module, and a dispatch and allocation module; The road monitoring module is used to collect road monitoring images, identify all vehicles, obtain road speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. The road monitoring module includes: a spatial conversion unit, an information access unit, and a network modeling unit; The spatial transformation unit is used to solve the camera's intrinsic parameters, including focal length, pitch angle, and yaw angle, through perspective geometric constraints. It establishes a self-calibration model based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and the actual position, and initializes the model parameters using road markings, vehicle trajectories, and road scenes. The information access unit is used to set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the operating conditions, divide the optimization range of the parameters according to the operating conditions, obtain real-time traffic flow data, use a distributed message queue to cache the real-time traffic flow, and build a dynamic information database for urban traffic. The network modeling unit is used to convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

[0024] The vehicle management module is used to establish a path planning architecture on the road network model, select paths based on road network traffic conditions and vehicle trajectories, set road segment rewards based on traffic density of each road segment, plan the optimal driving path through vehicle travel time and trajectory rewards, and distribute the results to each vehicle. The vehicle management module includes: a data association unit, a trajectory analysis unit, and a road segment planning unit; The data association unit is used to construct a database with traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the least support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. The trajectory analysis unit is used to process vehicle trajectories using a segmented embedding method, converting the trajectory embedding into a text feature representation, inputting it into an LLM for prediction, and embedding task requirements, map semantics, and environmental information into the model to determine the environmental input. The road segment planning unit uses environmental input as state, trajectory change as action, and the difference in average vehicle waiting time between time steps as reward. It calculates the reward function, outputs real-time road segment reward, and gives road segment reward related to the time spent to each node reached by the vehicle. The optimal path is calculated with the goal of maximizing reward and then distributed to the vehicle.

[0025] The accident analysis module is used to extract accident features from traffic police data, construct a road traffic accident dataset, output a prediction frame through a latent diffusion model based on the optical flow frame and video frame of the monitoring image, compare the pixel gray level difference between the prediction frame and the current frame, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster the principal component scores to obtain the average cluster comprehensive score, and determine the occurrence and type of the accident. The accident analysis module includes: an image comparison unit and an accident classification unit; The image comparison unit is used to extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. It diffuses the optical flow in potential spaces of two different scales, fuses motion information and appearance information, and fuses and decodes the output to generate prediction frames. It calculates the pixel grayscale difference between the prediction frames and the real frames to obtain a difference map, performs principal component analysis on the difference map, and obtains the principal component load matrix. The accident classification unit is used to project the difference map onto the principal component space to obtain the principal component score vector. The variance contribution rate of each principal component is calculated as a weight to weight the score, resulting in a weighted score vector. The principal component score vectors of historical accident samples are used to train a clustering model. The principal component score vector of the current abnormal event is compared with the mean of each cluster to calculate the similarity and determine the accident type.

[0026] The traffic focusing module is used to extract the spatiotemporal features of hotspot areas from road environment data to establish spatial connection relationships, identify hotspot areas based on hotspot analysis, calculate the spatiotemporal kernel density value of traffic accidents, analyze the distribution characteristics of traffic violations from a spatiotemporal dimension, and optimize patrol routes based on distribution characteristics to cover traffic hotspot areas. The traffic focusing module includes: a heat map connection unit and a feature optimization unit; The heat connection unit is used to establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, the local sum of the accident frequency with neighboring units is calculated and compared with the global expectation to calculate the hot spot and cold spot classification of the accident at different locations. The feature optimization unit is used to convert discrete hotspot density into a continuous probability density surface, generate a spatial kernel density map, identify spatiotemporally continuous clusters using spatiotemporal scanning statistics, cluster shape features and density features are improved by the DBSCAN algorithm, fine-grained division of cluster boundaries is performed to obtain road segment weights, and the maximum coverage path algorithm is used to plan patrol routes.

[0027] The scheduling and allocation module is used to connect media terminals of different standards to the scheduling platform, build an exchange service platform for command and dispatch, construct a two-dimensional directed network model, quantify the optimization degree of each facility and service for road accidents, construct an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

[0028] The scheduling and allocation module includes: a command platform unit and an element scheduling unit; The command platform unit is used to connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. The element scheduling unit is used to align multi-source data in time and space, classify accident risk levels, calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence through the weighted HITS algorithm as quantitative indicators of operational performance, and use a binomial Logit model to fit the influence weights of each influencing factor on the probability of traffic violation events.

[0029] The data-driven approach to road traffic risk management includes the following steps: Step S1. Collect road monitoring images, identify all vehicles, obtain road travel speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. Step S1 includes: Step S11. Solve the camera intrinsic parameters, including focal length, pitch angle and yaw angle, through perspective geometric constraints. Based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and actual position, establish a self-calibration model and initialize the model parameters using road markings, vehicle trajectories and road scenes. Step S12. Set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the working conditions, divide the optimization interval of the parameters according to the working conditions, obtain real-time traffic flow data, use the distributed message queue to cache the real-time traffic flow, and build a dynamic information database of urban traffic. Step S13. Convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

[0030] Step S2. Establish a path planning architecture on the road network model, set road segment rewards according to the traffic density of each road segment, plan the optimal driving route through vehicle travel time and trajectory rewards, and distribute it to each vehicle; Step S2 includes: Step S21. Construct a database using traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the lowest support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. Step S22. Process the vehicle trajectory using a segmented embedding method, convert the trajectory embedding into a text feature representation, input it into LLM for prediction, embed the task requirements, map semantics and environmental information into the model, determine the environmental input, the environmental input is the state, the trajectory change is the action, the difference in the average waiting time of the vehicle before and after is the reward, calculate the reward function, output the real-time road segment reward, and give the vehicle a road segment reward related to the time spent when it arrives at each node. Calculate the optimal path with the goal of maximizing the reward and send it to the vehicle.

[0031] Step S3. Extract accident features from traffic police data, construct a road traffic accident dataset, calculate the pixel grayscale difference between the predicted frame and the current frame of the monitoring image, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster to obtain the average cluster comprehensive score, and determine the accident type. Step S3 includes: Step S31. Extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. Diffusion is performed in potential spaces of two different scales to fuse motion information and appearance information. The output is then fused and decoded to generate prediction frames. The pixel grayscale difference between the prediction frames and the real frames is calculated to obtain a difference map. Principal component analysis is performed on the difference map to obtain the principal component load matrix. Step S32. Project the difference map onto the principal component space to obtain the principal component score vector. Calculate the variance contribution rate of each principal component as a weight to weight the score and obtain a weighted score vector. Use the principal component score vectors of historical accident samples to train a clustering model. Compare the principal component score vector of the current abnormal event with the mean of each cluster to calculate the similarity and determine the accident type.

[0032] Step S4. Identify hotspot areas based on hotspot analysis, establish spatial connectivity, calculate the spatiotemporal kernel density value of traffic accidents, and optimize patrol routes based on the distribution characteristics of traffic violations to cover traffic hotspot areas; Step S4 includes: Step S41. Establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, calculate the local sum of the accident frequencies with neighboring units and compare it with the global expectation. Calculate the hot spot and cold spot classification of the accident at different locations. Step S42. Convert discrete hotspot density into continuous probability density surface, generate spatial kernel density map, identify spatiotemporally continuous clusters using spatiotemporal scanning statistics, refine the cluster shape and density features by improving the DBSCAN algorithm, perform fine-grained division of cluster boundaries, obtain road segment weights, and use the maximum coverage path algorithm to plan patrol routes.

[0033] Step S5. Connect different media terminals to the dispatch platform, build an exchange service platform for command and dispatch, quantify the optimization degree of each facility and business for road accidents, build an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

[0034] Step S5 includes: Step S51. Connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. Step S52. Spatiotemporally align multi-source data, classify accident risk levels, and calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence using the weighted HITS algorithm as quantitative indicators of operational performance, and fit the influence weights of each influencing factor on the probability of traffic violation events using a binomial Logit model.

[0035] Example: Monitoring equipment is deployed at road nodes, video streams are preprocessed, a self-calibration model is established based on camera intrinsic and extrinsic parameters, the same vehicle in consecutive frames is associated, vehicle trajectories are extracted, instantaneous speed is calculated based on vehicle displacement and time difference in consecutive frames, traffic flow, density and average speed of each road segment are statistically analyzed, a traffic dataset is constructed, a road network model is obtained by abstracting the road network, local spatiotemporal features are captured through a neural operator network, a road network state prediction for the next time period is output, a reward is set based on the traffic flow density of the current road segment, and the optimal path is calculated using a dynamic programming algorithm. Historical traffic incident records are collected to construct an accident feature database. Video frames are predicted based on a potential diffusion model, pixel differences are calculated, principal component transformation is performed with variance contribution rate weighting, accidents are clustered and accident types are determined, road environment data is integrated, spatiotemporal features are extracted, hotspot areas are identified through hotspot analysis, a spatiotemporal weight matrix is ​​constructed, the spatiotemporal kernel density risk value of traffic events is estimated, and patrol routes are planned to maximize the total risk value covered while meeting the constraints of patrol duration and number of police officers.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0037] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A road traffic risk management method based on data analysis, characterized in that, The method includes the following steps: Step S1. Collect road monitoring images, identify all vehicles, obtain road travel speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. Step S2. Establish a path planning architecture on the road network model, set road segment rewards according to the traffic density of each road segment, plan the optimal driving route through vehicle travel time and trajectory rewards, and distribute it to each vehicle; Step S3. Extract accident features from traffic police data, construct a road traffic accident dataset, calculate the pixel grayscale difference between the predicted frame and the current frame of the monitoring image, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster to obtain the average cluster comprehensive score, and determine the accident type. Step S4. Identify hotspot areas based on hotspot analysis, establish spatial connectivity, calculate the spatiotemporal kernel density value of traffic accidents, and optimize patrol routes based on the distribution characteristics of traffic violations to cover traffic hotspot areas; Step S5. Connect different media terminals to the dispatch platform, build an exchange service platform for command and dispatch, quantify the optimization degree of each facility and business for road accidents, build an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

2. The road traffic risk management method based on data analysis according to claim 1, characterized in that: Step S1 includes: Step S11. Solve the camera intrinsic parameters, including focal length, pitch angle and yaw angle, through perspective geometric constraints. Based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and actual position, establish a self-calibration model and initialize the model parameters using road markings, vehicle trajectories and road scenes. Step S12. Set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the working conditions, divide the optimization interval of the parameters according to the working conditions, obtain real-time traffic flow data, use the distributed message queue to cache the real-time traffic flow, and build a dynamic information database of urban traffic. Step S13. Convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

3. The road traffic risk management method based on data analysis according to claim 2, characterized in that: Step S2 includes: Step S21. Construct a database using traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the lowest support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. Step S22. Process the vehicle trajectory using a segmented embedding method, convert the trajectory embedding into a text feature representation, input it into LLM for prediction, embed the task requirements, map semantics and environmental information into the model, determine the environmental input, the environmental input is the state, the trajectory change is the action, the difference in the average waiting time of the vehicle before and after is the reward, calculate the reward function, output the real-time road segment reward, give the vehicle a road segment reward related to the time spent when it reaches a node, calculate the optimal path with the goal of maximizing the reward, and send it to the vehicle.

4. The road traffic risk management method based on data analysis according to claim 3, characterized in that: Step S3 includes: Step S31. Extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. Diffusion is performed in potential spaces of two different scales to fuse motion information and appearance information. The output is then fused and decoded to generate prediction frames. The pixel grayscale difference between the prediction frames and the real frames is calculated to obtain a difference map. Principal component analysis is performed on the difference map to obtain the principal component load matrix. Step S32. Project the difference map onto the principal component space to obtain the principal component score vector. Calculate the variance contribution rate of each principal component as a weight to weight the score and obtain a weighted score vector. Use the principal component score vectors of historical accident samples to train a clustering model. Compare the principal component score vector of the current abnormal event with the mean of each cluster to calculate the similarity and determine the accident type.

5. The road traffic risk management method based on data analysis according to claim 4, characterized in that: Step S4 includes: Step S41. Establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, calculate the local sum of the accident frequencies with neighboring units and compare it with the global expectation. Calculate the hot spot and cold spot classification of the accident at different locations. Step S42. Convert the discrete hotspot density into a continuous probability density surface to generate a spatial kernel density map. Use spatiotemporal scanning statistics to identify spatiotemporally continuous clusters. Improve the DBSCAN algorithm to cluster shape and density features. Perform fine-grained division of cluster boundaries to obtain road segment weights. Use the maximum coverage path algorithm to plan patrol routes. Step S5 includes: Step S51. Connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. Step S52. Spatiotemporally align multi-source data, classify accident risk levels, and calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence using the weighted HITS algorithm as quantitative indicators of operational performance, and fit the influence weights of each influencing factor on the probability of traffic violation events using a binomial Logit model.

6. A road traffic risk management and control system based on data analysis, characterized in that: The system includes the following modules: road monitoring module, vehicle management module, accident analysis module, traffic focusing module, and dispatching and allocation module; The road monitoring module is used to collect road monitoring images, identify all vehicles, obtain road speed based on real-time traffic flow, and establish a road network model using road travel time as spatial distance. The vehicle management module is used to establish a path planning architecture on the road network model, select paths based on road network traffic conditions and vehicle trajectories, set road segment rewards based on traffic density of each road segment, plan the optimal driving path through vehicle travel time and trajectory rewards, and distribute the results to each vehicle. The accident analysis module is used to extract accident features from traffic police data, construct a road traffic accident dataset, output a prediction frame through a latent diffusion model based on the optical flow frame and video frame of the monitoring image, compare the pixel gray level difference between the prediction frame and the current frame, process the variance contribution rate through the principal component load matrix corresponding to each dataset, calculate the principal component score of each accident, cluster the principal component scores to obtain the average cluster comprehensive score, and determine the occurrence and type of the accident. The traffic focusing module is used to extract the spatiotemporal features of hotspot areas from road environment data to establish spatial connection relationships, identify hotspot areas based on hotspot analysis, calculate the spatiotemporal kernel density value of traffic accidents, analyze the distribution characteristics of traffic violations from a spatiotemporal dimension, and optimize patrol routes based on distribution characteristics to cover traffic hotspot areas. The scheduling and allocation module is used to connect media terminals of different standards to the scheduling platform, build an exchange service platform for command and dispatch, construct a two-dimensional directed network model, quantify the optimization degree of each facility and service for road accidents, construct an element collaboration strength network based on quantitative indicators, identify element strength levels, and conduct command and dispatch according to element failure levels.

7. The road traffic risk management system based on data analysis according to claim 6, characterized in that: The road monitoring module includes: a spatial conversion unit, an information access unit, and a network modeling unit; The spatial transformation unit is used to solve the camera's intrinsic parameters, including focal length, pitch angle, and yaw angle, through perspective geometric constraints. It establishes a self-calibration model based on the spatial coordinate relationship between pixel coordinates in the monitoring equipment and the actual position, and initializes the model parameters using road markings, vehicle trajectories, and road scenes. The information access unit is used to set the minimum calibration accuracy of the self-calibration model, group the initial conditions to obtain the operating conditions, divide the optimization range of the parameters according to the operating conditions, obtain real-time traffic flow data, use a distributed message queue to cache the real-time traffic flow, and build a dynamic information database for urban traffic. The network modeling unit is used to convert vehicle trajectories into vectors, identify the contextual relationships of trajectories through a bidirectional long short-term memory network, obtain triples from entity labelers, construct a traffic dataset, and fuse the embedded vectors of weather, traffic flow, and road segment time slice entities with the features of the traffic dataset for nonlinear modeling.

8. The road traffic risk management system based on data analysis according to claim 7, characterized in that: The vehicle management module includes: a data association unit, a trajectory analysis unit, and a road segment planning unit; The data association unit is used to construct a database with traffic flow data, calculate the frequent sets between different databases, retain the common frequent set parts, calculate the support of other parts in the database, retain the frequent set with the least support as the training basis, divide the time domain into sub-intervals, train a neural operator network for each interval, and capture local features. The trajectory analysis unit is used to process vehicle trajectories using a segmented embedding method, converting the trajectory embedding into a text feature representation, inputting it into an LLM for prediction, and embedding task requirements, map semantics, and environmental information into the model to determine the environmental input. The road segment planning unit uses environmental input as state, trajectory change as action, and the difference in average vehicle waiting time between time steps as reward. It calculates the reward function, outputs real-time road segment reward, and gives road segment reward related to the time spent to each node reached by the vehicle. The optimal path is calculated with the goal of maximizing reward and then distributed to the vehicle.

9. The road traffic risk management system based on data analysis according to claim 8, characterized in that: The accident analysis module includes: an image comparison unit and an accident classification unit; The image comparison unit is used to extract video frame sequences before, at the moment of, and after the accident, and calculate the corresponding optical flow frames. It diffuses the optical flow in potential spaces of two different scales, fuses motion information and appearance information, and fuses and decodes the output to generate prediction frames. It calculates the pixel grayscale difference between the prediction frames and the real frames to obtain a difference map, performs principal component analysis on the difference map, and obtains the principal component load matrix. The accident classification unit is used to project the difference map onto the principal component space to obtain the principal component score vector. The variance contribution rate of each principal component is calculated as a weight to weight the score, resulting in a weighted score vector. The principal component score vectors of historical accident samples are used to train a clustering model. The principal component score vector of the current abnormal event is compared with the mean of each cluster to calculate the similarity and determine the accident type.

10. The road traffic risk management system based on data analysis according to claim 9, characterized in that: The traffic focusing module includes: a heat map connection unit and a feature optimization unit; The heat connection unit is used to establish a spatiotemporal model of the accident based on the spatiotemporal proximity and spatiotemporal weight matrix. For each spatiotemporal unit, the local sum of the accident frequency with neighboring units is calculated and compared with the global expectation to calculate the hot spot and cold spot classification of the accident at different locations. The feature optimization unit is used to convert discrete hotspot density into continuous probability density surface, generate spatial kernel density map, identify spatiotemporally continuous clusters using spatiotemporal scanning statistics, cluster shape features and density features by improving DBSCAN algorithm, perform fine-grained division of cluster boundaries, obtain road segment weights, and use maximum coverage path algorithm to plan patrol routes. The scheduling and allocation module includes: a command platform unit and an element scheduling unit; The command platform unit is used to connect fixed surveillance cameras, police mobile terminals, public terminals and information release terminals to the command platform. The platform collects multi-source data and distributes it to each business module, and sends command and dispatch instructions to the corresponding terminals. The element scheduling unit is used to align multi-source data in time and space, classify accident risk levels, calculate the probability of an accident based on the conditional probability between each influencing factor, calculate facility support and business dependence through the weighted HITS algorithm as quantitative indicators of operational performance, and use a binomial Logit model to fit the influence weights of each influencing factor on the probability of traffic violation events.

Images