Traffic accident investigation method, device and system

Abstract

The invention relates to the technical field of unmanned aerial vehicles, in particular to a traffic accident investigation method, device and system, and the method comprises the steps: obtaining vehicle driving data, judging the road condition according to the vehicle driving data, and obtaining the road vehicle driving state information; obtaining vehicle position information, adjusting the vehicle position information according to the accident position adjustment prediction model to obtain destination information, and planning a route according to the destination information to obtain route information; planning and dispatching the unmanned aerial vehicle according to the route information; unmanned aerial vehicle feedback information is obtained, the accident position adjustment prediction model is optimized according to the unmanned aerial vehicle feedback information, the vehicle position information, the vehicle driving data and the real-time feature information, and the traffic accident investigation efficiency of the unmanned aerial vehicle is improved.

Description

Technical Field

[0001] This application relates to the technical field of unmanned aerial vehicles (UAVs), and in particular to a method, apparatus and system for investigating and handling traffic accidents. Background Technology

[0002] With the increasing maturity and widespread adoption of drone technology, it has demonstrated enormous application potential across numerous industries. In traffic accident handling, drones, with their advantages of rapid response, flexibility, comprehensive perspective, and independence from ground traffic restrictions, have gradually become an important tool for assisting in on-site investigation. Traditional accident handling requires traffic police officers to drive to the scene, often resulting in delays due to traffic congestion. Furthermore, close-range investigations pose safety risks and increase traffic pressure. Currently, existing technologies utilize drones for preliminary investigations. Based on roadside sensing devices or initial alarm information, after a remote assessment by personnel indicating a potential accident, the operator plans a flight path for the drone, directing it to the suspected scene to conduct aerial photography and transmit the video or images back to the command center. This method reduces the time spent by traffic police on their initial arrival at the scene and provides a global perspective.

[0003] However, the current technology has a low level of intelligence and automation, with core processes heavily reliant on human intervention, failing to fully realize the effectiveness of drone systems. Currently, the entire process relies on human intervention to review and assess initial information, such as congestion detection and emergency calls. Even after a drone arrives on-site, its task is limited to taking pictures; subsequent processing still requires significant effort from personnel at the rear or those arriving later to analyze and make decisions. This essentially shifts some on-site work remotely, without fundamentally reducing the total amount of human resources required, resulting in low work efficiency. These issues need to be addressed. Summary of the Invention

[0004] To more intelligently respond to traffic accidents and improve the efficiency of traffic accident investigation using drones, this application provides a traffic accident investigation method, apparatus, and system, employing the following technical solution: Firstly, this application provides a method for investigating and handling traffic accidents, including: Acquire vehicle driving data, assess road conditions based on the vehicle driving data, and obtain road vehicle driving status information; Obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route according to the destination information to obtain route information; Plan and schedule drones based on flight route information; Obtain feedback information from drones, and optimize the accident location adjustment prediction model based on drone feedback information, vehicle location information, vehicle driving data, and real-time feature information.

[0005] Preferably, the vehicle driving data includes visual frame data, millimeter-wave radar data, time sequence information, vehicle distance visual recognition data, and vehicle speed recognition data; The specific steps for judging road conditions based on vehicle driving data to obtain road vehicle driving status information are as follows: Visual frame data, millimeter-wave radar data, temporal information, vehicle distance visual recognition data, and vehicle speed recognition data are preprocessed and feature fused to obtain fused feature information; An accident determination information set is constructed based on the fused feature information. The road conditions are then determined based on the accident determination information set to obtain road vehicle driving status information.

[0006] Preferably, the specific steps for obtaining vehicle location information, adjusting the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and planning a route based on the destination information to obtain route information are as follows: Obtain vehicle location information, and when the road vehicle driving status information is abnormal, obtain real-time feature information. Then, use the accident position adjustment prediction model to estimate the deviation of vehicle driving data, vehicle location information and real-time feature information to obtain position adjustment information. The vehicle's location information is adjusted based on the location adjustment information to obtain the destination information. The route is then planned based on the destination information to obtain the route information.

[0007] Preferred options also include: Road data acquisition devices collect road vehicle driving data and vehicle location information.

[0008] Preferred options also include: The drone matches flight command information with route information; Acquire accident scene information, identify the accident location from the accident scene information, and obtain the accident location identification results and corresponding on-site handling instructions; Obtain information about the vehicles involved in the accident, identify the accident details and mark the accident points to obtain the investigation results.

[0009] Preferably, the specific steps of the drone are as follows: Drones were used to acquire on-site imagery. Based on the preset first visual model, the accident location is identified from the scene map information to obtain the accident location identification result; Match the corresponding on-site handling instructions based on vehicle location information and accident location identification results; Obtain global view map information, precise view map information, detailed view map information, and supplementary view map information; Based on the preset second vision model, accident situation identification and accident point marking are performed on global view map information, precise view map information, detailed view map information and supplementary view map information to obtain exploration result information.

[0010] Preferably, the drone feedback information includes accident location identification results and survey results.

[0011] Secondly, this application provides a traffic accident investigation and management platform, including: The driving status judgment module is used to acquire vehicle driving data, judge road conditions based on vehicle driving data, and obtain road vehicle driving status information. The route planning module is used to obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route information based on the destination information to obtain route information. The scheduling module is used to plan and schedule drones based on flight route information; The iteration module is used to acquire feedback information from the UAV and optimize the accident location adjustment prediction model based on the UAV feedback information, vehicle location information, vehicle driving data, and real-time feature information.

[0012] Thirdly, this application provides a traffic accident investigation and handling system, wherein the memory stores computer programs, including the traffic accident investigation and handling management platform as described above.

[0013] Preferably, road data acquisition devices and drones are also included.

[0014] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: This application monitors and judges road conditions by acquiring vehicle driving data. In the event of abnormal road conditions, it adjusts the initially acquired vehicle position information using an accident position adjustment prediction model to obtain adjusted destination information. Based on the destination information, it plans a flight path to obtain flight path information for output to the UAV to perform the task. The UAV is dispatched through the flight path information. After receiving the planned flight path information, the UAV matches the flight command information and performs the corresponding flight actions. After the UAV records the accident information on site, it obtains UAV feedback information. Based on the UAV feedback information, vehicle position information, vehicle driving data, and real-time feature information, the accident position adjustment prediction model is optimized to improve the prediction accuracy of the accident position adjustment prediction model, reduce human intervention, enable more intelligent response to traffic accidents, and improve the efficiency of UAV in handling traffic accidents. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a traffic accident investigation method as described in an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of the modules of the traffic accident investigation and management platform described in the embodiments of this application.

[0017] Figure 3 This is a schematic diagram of the modules of the traffic accident investigation system described in the embodiments of this application.

[0018] Explanation of reference numerals in the attached figures: 1. Road data acquisition device; 2. Traffic accident control platform; 21. Driving status judgment module; 22. Route planning module; 23. Dispatch module; 24. Iteration module; 3. Unmanned aerial vehicle (UAV). Detailed Implementation

[0019] The following combination Figures 1-3 The present application will be described in further detail below. The terminology used in the embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting.

[0020] Reference Figure 1 The method for handling traffic accidents involved in this application includes: Step S1: Obtain vehicle driving data, judge the road conditions based on the vehicle driving data, and obtain road vehicle driving status information; Step S2: Obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route according to the destination information to obtain route information; Step S3: Plan and schedule drones based on flight route information; Step S4: Obtain drone feedback information, and optimize the accident location adjustment prediction model based on drone feedback information, vehicle location information, vehicle driving data, and real-time feature information.

[0021] Specifically, this application provides a traffic accident investigation and handling method to improve the intelligence level of drones in responding to traffic accidents. This method is applied to a traffic accident investigation and handling management platform, i.e., the management platform. By acquiring vehicle driving data, road conditions are monitored and assessed. In cases of abnormal road conditions, an accident location adjustment prediction model is invoked to adjust the initially acquired vehicle location information, making the vehicle location more accurate and obtaining adjusted destination information. Based on the destination information, a flight path is planned, resulting in flight path information for output to the drone for task execution. The drone is then scheduled to perform the task using this flight path information. After receiving the planned flight path information, the drone matches the flight command information and executes the corresponding flight actions. Upon arriving at the accident scene, the drone first identifies the accident location, accurately determines the accident location, performs on-site maintenance and handling operations, and then surveys and records the accident vehicles at the accident location. The recorded accident vehicle information is then used to identify and mark specific accident points, obtaining feedback information from the drone. After the drone records the accident information at the scene, the drone feedback information is obtained. Based on the drone feedback information, vehicle location information, vehicle driving data and real-time feature information, the accident location adjustment prediction model is optimized to improve the prediction accuracy of the accident location adjustment prediction model, reduce human intervention, enable more intelligent response to traffic accidents, and improve the efficiency of drone traffic accident investigation.

[0022] As one implementation method, it also includes: Road data acquisition devices collect road vehicle driving data and vehicle location information.

[0023] Specifically, the road data acquisition device is a multi-source monitoring system constructed by linking electronic police or checkpoint equipment with visual sensors and millimeter-wave radar to collect road vehicle data in real time.

[0024] Vehicle driving data includes visual frame data, millimeter-wave radar data, time series information, vehicle distance visual recognition data, and vehicle speed recognition data, while vehicle location information specifically refers to the vehicle's real-time latitude and longitude coordinates.

[0025] As one implementation method, vehicle driving data includes visual frame data, millimeter-wave radar data, timing information, vehicle distance visual recognition data, and vehicle speed recognition data. The specific steps for judging road conditions based on vehicle driving data to obtain road vehicle driving status information are as follows: Step S11: Preprocess and fuse the visual frame data, millimeter-wave radar data, temporal information, vehicle distance visual recognition data, and vehicle speed recognition data to obtain fused feature information; Step S12: Construct an accident determination information set based on the fused feature information, and determine the road conditions based on the accident determination information set to obtain road vehicle driving status information.

[0026] Specifically, this application embodiment, through three stages of multi-parameter acquisition, feature fusion, and decision-making, adds vehicle distance recognition and vehicle speed recognition on the basis of environmental perception and feature fusion.

[0027] The input layer data includes visual frame data, millimeter-wave radar data, temporal information, vehicle distance visual recognition data, and vehicle speed recognition data. Visual frame data consists of RGB or infrared related data, used for vehicle contour recognition, vehicle distance visual calculation, and on-site trace capture. Millimeter-wave radar data provides precise vehicle distance, real-time speed, and relative position data. Temporal information specifically constructs a 10-frame temporal window to record continuous changes in vehicle driving status. Vehicle distance visual recognition data, based on the principle of binocular vision ranging, calculates the parallax between the images acquired by the left and right cameras, combined with camera intrinsic calibration parameters, to obtain the actual distance between vehicles. The calculation formula is D=(f×B) / d, where D is the vehicle distance, f is the camera focal length, B is the binocular camera baseline length, and d is the parallax. Vehicle speed recognition data is obtained by directly acquiring the instantaneous speed of the vehicle through millimeter-wave radar. At the same time, it is combined with the pixel displacement of the vehicle in the visual frame between consecutive frames. The actual driving speed of the vehicle is calculated based on the perspective projection transformation model. The formula is v=(Δx×S) / (Δt×P), where v is the vehicle speed, Δx is the vehicle pixel displacement between consecutive frames, S is the ratio coefficient of the actual road distance to the pixel distance, Δt is the frame interval time, and P is the camera pixel scale.

[0028] This application performs preprocessing when acquiring vehicle driving data. The preprocessing layer employs adaptive rain or fog removal algorithms for adverse weather conditions such as rain, snow, and fog. An improved algorithm combining dark channel prior and gamma correction is used to preprocess visual images, improving image clarity. Kalman filtering is applied to millimeter-wave radar data to filter random noise and ensure the stability of vehicle speed and distance data. Cross-calibration is performed on visual distance recognition data and radar ranging data to eliminate single-sensor errors and obtain accurate vehicle distance values. Moving average filtering is applied to vehicle speed data to remove instantaneous fluctuations and obtain stable vehicle speeds.

[0029] After preprocessing, the weights of each modality feature are dynamically assigned based on an improved attention mechanism. The fused features include vehicle visual features F_visual, radar distance or speed features F_radar, temporal features F_temporal, vehicle distance features F_distance, and vehicle speed features F_speed. The fusion formula is F_fusion = α·F_visual + β·F_radar + γ·F_temporal + δ·F_distance + ε·F_speed, where α + β + γ + δ + ε = 1. The weight coefficients are adjusted in real time through an adaptive learning algorithm.

[0030] In one embodiment, under normal weather and unobstructed conditions, α=0.3, β=0.2, γ=0.1, δ=0.2, and ε=0.2.

[0031] In another embodiment, under adverse weather or rain, snow and fog scenarios, the radar feature weight β is increased to 0.35, the vehicle distance and vehicle speed feature weights δ and ε are each increased to 0.25, α is adjusted to 0.1, and γ is kept at 0.05 to resist visual feature distortion.

[0032] In another embodiment, under the occlusion scenario during morning and evening rush hours, the weight of the temporal feature γ is increased to 0.2, the weight of the vehicle distance feature δ is increased to 0.3, ε=0.2, α=0.15, β=0.15, and the recognition is assisted by the continuous driving status of vehicles and the change of distance.

[0033] This application constructs an accident determination information set based on fusion features and combines a confidence leveling mechanism to achieve accurate decision-making.

[0034] A high-confidence determination, defined as a confidence level ≥ 0.85, is made if any of the following conditions are met: The vehicle speed suddenly drops to 0 km / h, and the distance between adjacent vehicles is ≤ 1 meter (less than the safe distance threshold), while visual features detect collision deformation or traces on the vehicle; or multi-frame time-series data shows the vehicle speed is 0 km / h for more than 3 consecutive frames, and the distance between vehicles remains less than the safe distance, while radar detects no change in the relative position of the vehicles.

[0035] For medium confidence level determination (0.6 ≤ confidence level < 0.85), inter-frame verification is required if the following conditions are met: vehicle speed drops to 0-5 km / h, distance between adjacent vehicles is 1-3 meters, and no clear collision marks are visible. A 3-frame verification process is initiated. If the above conditions are met for 3 consecutive frames and the distance between vehicles does not increase, the determination is upgraded to a traffic accident determination. If the distance between vehicles increases or vehicles resume driving during this period, the determination is normal parking or slow movement.

[0036] Low confidence level judgment, i.e. confidence level < 0.6, where the vehicle speed is normal, the distance between vehicles meets safety standards, or only a single frame of data is abnormal, is directly filtered out and judged as a non-accident scenario.

[0037] This application significantly improves the accuracy and robustness of accident recognition through the synergistic enhancement of vehicle distance visual recognition and vehicle speed recognition. Real-world testing has verified that the accident false alarm rate is reduced by more than 85% under adverse weather conditions, the false alarm rate is controlled below 5% in peak-hour obstruction scenarios, and the inference latency is <70ms, meeting the real-time requirements of edge devices. It is suitable for 24-hour monitoring in various scenarios such as highways, urban main roads, and urban secondary roads.

[0038] As one implementation method, the specific steps for obtaining vehicle location information, adjusting the vehicle location information according to an accident location adjustment prediction model to obtain destination information, and planning a route based on the destination information are as follows: Step S21: Obtain vehicle location information. When the road vehicle driving status information is abnormal, obtain real-time feature information. Use the accident location adjustment prediction model to estimate the deviation of vehicle driving data, vehicle location information and real-time feature information to obtain location adjustment information. Step S22: Adjust the vehicle location information according to the location adjustment information to obtain the destination information, and plan the route according to the destination information to obtain the route information.

[0039] Specifically, in this embodiment, after receiving precise destination information, the control platform initiates a full-process scheduling process including positioning calibration, drone matching, and route optimization, with route planning and drone assignment proceeding in parallel. The control platform implements route planning functionality. By combining a 3D model of the urban traffic network, it performs geographic information matching on the uploaded latitude and longitude range, eliminating interference factors such as building obstructions and road forks, locking the physical coordinates of the accident site, and generating target locations that the drone can accurately reach.

[0040] The intelligent drone dispatch system searches the drone status database within the scheduling range in real time and filters idle devices that meet certain conditions. Specifically, drones with sufficient remaining battery power to support round-trip flights and on-site operations are selected, typically with a redundancy of ≥30%; payload compatibility requires the drone to be equipped with a high-definition camera, infrared module, and loudspeaker; there should be no fault alarms, and the drone should be within a legal flight area; multiple qualified drones are prioritized using distance-first and state-optimal algorithms to quickly complete the dispatch.

[0041] This application specifically describes a dual-mode flight path planning. In the conventional mode, staff mark the accident core area and surrounding safety boundaries on a satellite map based on calibrated latitude and longitude points, manually adjust the flight path to avoid obstacles, and directly issue flight commands. The automatic mode uses the A* optimization algorithm, integrating real-time meteorological data such as wind speed, visibility, no-fly zone data, and UAV performance parameters to generate an optimal flight path with a path deviation of ≤ ±2 meters.

[0042] If an incident remains unresolved within a specified timeframe (e.g., ≤5 minutes on urban roads, ≤3 minutes on highways), the dispatch strategy is automatically upgraded. Priority is given to dispatching the nearest drone, while the flight path planning calculation cycle is shortened and flight altitude is adjusted to improve arrival efficiency, thereby reducing response delays.

[0043] When the drone acquires the route information, it matches the corresponding flight command information with the route information, so that the drone can perform flight operations according to certain flight commands until it reaches the estimated vehicle position and waits for the drone to perform the next operation.

[0044] After the drone completes the survey and processing of the accident site, this application will feed back the obtained process information or result information to the management and control platform, laying the foundation for the updating and iteration of the management and control platform.

[0045] After acquiring vehicle driving data, vehicle location information, accident location identification results, and real-time feature information, the control platform can calculate the positional deviation of the UAV from the initial arrival at the accident scene and the precise accident location using the vehicle location information and accident location identification results. This application adjusts the initially obtained vehicle location information using an accident location adjustment prediction model to make the vehicle location information more accurate.

[0046] The training process of the accident location adjustment prediction model specifically uses vehicle driving data as a disturbance factor. It may also include, but is not limited to, acquiring several disturbance factors from road data acquisition devices, namely real-time feature information, such as environmental features of the accident location, accident time features, the status and location of monitoring equipment, features of the preliminary accident description, and historical data of the above features. The disturbance factors are used as initial features, and the positional deviation between vehicle location information and accident location identification results is used as a label. Methods such as linear regression, decision trees, and neural networks are selected to predict the deviation.

[0047] This application deploys the trained model into a management and control platform. When a new preliminary accident report arrives, the platform not only provides an initial predicted location based on monitoring data, but also uses this model to predict a possible deviation based on the characteristics of the current accident. Then, the initial predicted location is added to the predicted deviation to obtain a corrected predicted location. This corrected location is used to plan the UAV flight path and output flight path information. This application's embodiments improve the accuracy of initial accident location prediction through feedback data training, thereby reducing the time spent by the UAV adjusting its location after arriving at the accident scene and improving the efficiency of UAV-based traffic accident investigation.

[0048] As one implementation method, it also includes: The drone matches flight command information with route information; Acquire accident scene information, identify the accident location from the accident scene information, and obtain the accident location identification results and corresponding on-site handling instructions; Obtain information about the vehicles involved in the accident, identify the accident details and mark the accident points to obtain the investigation results.

[0049] Specifically, in this embodiment of the application, the UAV is a device for performing investigation and reconnaissance tasks. After receiving the planned route information, it matches the flight command information to implement the corresponding flight actions. Upon arriving at the accident scene, it first identifies the accident location, determines the accurate accident location, performs on-site maintenance and handling operations, and then investigates and records the accident vehicles at the accident location. The accident information acquisition module identifies and marks the specific accident points on the recorded accident vehicle information to obtain the investigation result information.

[0050] As one implementation method, it also includes: The drone matches flight command information with route information; Acquire accident scene information, identify the accident location from the accident scene information, and obtain the accident location identification results and corresponding on-site handling instructions; Obtain information about the vehicles involved in the accident, identify the accident details and mark the accident points to obtain the investigation results.

[0051] Specifically, the UAV of this application enables the acquisition of on-site map information, identification of accident locations, and generation of on-site handling instructions.

[0052] Since the final destination in the flight path information output by the control platform is preliminary information, it cannot completely plan the drone to the vicinity of the accident vehicle. This application obtains the scene map information taken by the drone after it reaches the destination, and uses the established first vision model to identify the accident location from the scene map information to obtain the accident location identification result.

[0053] As one implementation method, the specific steps of using a drone are as follows: Drones were used to acquire on-site imagery. Based on the preset first visual model, the accident location is identified from the scene map information to obtain the accident location identification result; Match the corresponding on-site handling instructions based on vehicle location information and accident location identification results; Obtain global view map information, precise view map information, detailed view map information, and supplementary view map information; Based on the preset second vision model, accident situation identification and accident point marking are performed on global view map information, precise view map information, detailed view map information and supplementary view map information to obtain exploration result information.

[0054] The drone's feedback information includes accident location identification results and survey results.

[0055] Specifically, the first visual model in this application is a dedicated computer vision model designed for accurately locating the core accident occurrence area in complex road scene images. In a panoramic image containing roads, multiple vehicles, roadside facilities, etc., it automatically frames or segments the core area directly related to the accident. The model learns long-distance dependencies between different regions in the image through self-attention or a visual Transformer module. This enables it to understand causal relationships, such as the brake lights illuminating on a distant vehicle and the abnormal stopping of a nearby vehicle, thereby more accurately locating the source of the accident, not limited to the abnormal object itself, and reducing human judgment.

[0056] After obtaining the accident location identification results, the camera calibration was used to establish the correspondence between pixel coordinates and light direction in the camera coordinate system, including focal length (fx, fy), optical center (cx, cy), lens distortion coefficient, etc., so as to calculate the coordinates of the accident location.

[0057] The accident location identification result specifically includes further vehicle location latitude and longitude information. One embodiment involves processing the data within the current UAV to calculate the distance and orientation between the vehicle location information and the accident location identification result, and matching this information to obtain on-site processing instructions for the UAV to travel from the vehicle location information to the accident location identification result.

[0058] The on-site handling instructions include, but are not limited to, flight commands for the drone from one location to another, but also instructions corresponding to on-site handling steps. These include safety announcements and environmental monitoring, guiding the drone to a safe altitude, such as 5-10 meters, which can be adaptively adjusted according to on-site traffic flow, and then continuously playing safety prompts through a microphone device. At the same time, on-site environmental data, such as visibility and whether there are any people loitering, are collected in real time and fed back to the control platform to assist in adjusting algorithm parameters.

[0059] The information obtained in this application embodiment is specifically obtained by drones through multi-dimensional evidence collection. This includes a global perspective: a 720° panoramic view directly above the accident site, covering the scene and a 50-meter radius around it; a precise perspective: a vertical top-down view, restoring the relative positions of vehicles at a 1:100 scale; a detailed perspective: a 45° tilted top-down view of the four corners of the quadrilateral area of ​​the accident zone, clearly capturing brake marks and vehicle collision traces; and a supplementary perspective: an infrared image, adapted for low-visibility scenarios such as nighttime, rain, snow, and fog, assisting in the reconstructing of vehicle distance and speed.

[0060] In this embodiment, standardized accident symbols, such as brake tire marks, bloodstains, and vehicle collision point markings, are overlaid on the image that needs to be fed back to the control platform. At the same time, based on the visual recognition data of vehicle distance, the initial distance between the vehicles involved and the change in distance after the collision are marked, and a visual survey report is generated.

[0061] The second visual model in this application is a pre-trained object detection model, such as YOLO or Faster R-CNN, which automatically locates and identifies key elements in the image, such as the vehicle involved, brake tracks, bloodstains, and debris. It delineates the pixel-level boundaries of the brake tracks and vehicle outlines, providing a foundation for subsequent measurement, annotation, and iteration.

[0062] Reference Figure 2 This application provides a traffic accident investigation and management platform, comprising: The driving status judgment module is used to acquire vehicle driving data, judge road conditions based on vehicle driving data, and obtain road vehicle driving status information. The route planning module is used to obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route information based on the destination information to obtain route information. The scheduling module is used to plan and schedule drones based on flight route information; The iteration module is used to acquire feedback information from the UAV and optimize the accident location adjustment prediction model based on the UAV feedback information, vehicle location information, vehicle driving data, and real-time feature information.

[0063] Reference Figure 3 This application provides a traffic accident investigation and handling system, including the traffic accident investigation and handling management platform as described above.

[0064] One implementation method also includes road data acquisition devices and drones.

[0065] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device and product described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0066] In the several embodiments provided in this application, it should be understood that the disclosed methods, systems, apparatus and program products can be implemented in other ways.

[0067] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0068] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for investigating and handling traffic accidents, characterized in that, include: Acquire vehicle driving data, assess road conditions based on the vehicle driving data, and obtain road vehicle driving status information; Obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route according to the destination information to obtain route information; Plan and schedule drones based on flight route information; Obtain feedback information from drones, and optimize the accident location adjustment prediction model based on drone feedback information, vehicle location information, vehicle driving data, and real-time feature information.

2. The method for investigating and handling traffic accidents according to claim 1, characterized in that, The vehicle driving data includes visual frame data, millimeter-wave radar data, time sequence information, vehicle distance visual recognition data, and vehicle speed recognition data; The specific steps for judging road conditions based on vehicle driving data to obtain road vehicle driving status information are as follows: Visual frame data, millimeter-wave radar data, temporal information, vehicle distance visual recognition data, and vehicle speed recognition data are preprocessed and feature fused to obtain fused feature information; An accident determination information set is constructed based on the fused feature information. The road conditions are then determined based on the accident determination information set to obtain road vehicle driving status information.

3. The method for investigating and handling traffic accidents according to claim 1, characterized in that, The specific steps for obtaining vehicle location information, adjusting the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and planning a route based on the destination information to obtain route information are as follows: Obtain vehicle location information, and when the road vehicle driving status information is abnormal, obtain real-time feature information. Then, use the accident position adjustment prediction model to estimate the deviation of vehicle driving data, vehicle location information and real-time feature information to obtain position adjustment information. The vehicle's location information is adjusted based on the location adjustment information to obtain the destination information. The route is then planned based on the destination information to obtain the route information.

4. The method for investigating and handling traffic accidents according to claim 1, characterized in that, Also includes: Road data acquisition devices collect road vehicle driving data and vehicle location information.

5. The method for investigating and handling traffic accidents according to claim 1, characterized in that, Also includes: The drone matches flight command information with route information; Acquire accident scene information, identify the accident location from the accident scene information, and obtain the accident location identification results and corresponding on-site handling instructions; Obtain information about the vehicles involved in the accident, identify the accident details and mark the accident points to obtain the investigation results.

6. The method for investigating and handling traffic accidents according to claim 5, characterized in that, The specific steps for using the drone are as follows: Drones were used to acquire on-site imagery. Based on the preset first visual model, the accident location is identified from the scene map information to obtain the accident location identification result; Match the corresponding on-site handling instructions based on vehicle location information and accident location identification results; Obtain global view map information, precise view map information, detailed view map information, and supplementary view map information; Based on the preset second vision model, accident situation identification and accident point marking are performed on global view map information, precise view map information, detailed view map information and supplementary view map information to obtain exploration result information.

7. The method for investigating and handling traffic accidents according to claim 5, characterized in that, The information fed back by the drone includes the accident location identification results and the survey results.

8. A traffic accident investigation and management platform, characterized in that, include: The driving status judgment module is used to acquire vehicle driving data, judge road conditions based on vehicle driving data, and obtain road vehicle driving status information. The route planning module is used to obtain vehicle location information, adjust the vehicle location information according to the accident location adjustment prediction model to obtain destination information, and plan the route information based on the destination information to obtain route information. The scheduling module is used to plan and schedule drones based on flight route information; The iteration module is used to acquire feedback information from the UAV and optimize the accident location adjustment prediction model based on the UAV feedback information, vehicle location information, vehicle driving data, and real-time feature information.

9. A traffic accident investigation and handling system, characterized in that, This includes the traffic accident investigation and management platform as described in claim 9.

10. The traffic accident investigation and handling system according to claim 9, characterized in that, It also includes road data collection devices and drones.

Images