Virtual simulation dynamic path test method based on accident scene fixed track

Abstract

The invention belongs to the technical field of automatic driving, and relates to a virtual simulation dynamic path test method based on a fixed trajectory of an accident scene, which comprises the following steps: S1, collecting and preprocessing multi-source data, analyzing the data to generate point cloud information and visual information, and extracting vehicle parameter information, vehicle behavior information, environment information and traffic participant abnormal behavior information; the vehicle behavior information comprises a fixed track; s2, a virtual simulation scene is constructed, a virtual three-dimensional scene model is constructed according to the point cloud data and the visual information, and parameters of the virtual three-dimensional scene model are set according to the vehicle parameter information, the vehicle behavior information, the environment information, the traffic participant abnormal behavior information and the traffic rules; s3, testing by taking the fixed track as a reference track, and performing dynamic path adjustment by utilizing a path planning algorithm in an automatic driving system during testing; and S4, analyzing the test data, evaluating from multiple dimensions, generating an analysis report, and optimizing suggestions. The problem of inaccurate test results is solved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and in particular relates to a virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario. Background Technology

[0002] With the rapid development of autonomous driving technology, virtual simulation testing has become a crucial step in verifying system performance and ensuring safety and reliability. Compared to real-vehicle road testing, virtual simulation can efficiently reproduce complex traffic scenarios in a controlled environment, covering extreme and dangerous conditions, significantly reducing testing costs and risks. Simultaneously, simulation testing supports large-scale parallel verification, providing an efficient platform for the iterative optimization of autonomous driving algorithms.

[0003] Currently, path planning in autonomous driving virtual simulation testing mainly relies on dynamic algorithm generation. This means the test vehicle autonomously decides its path based on preset scenarios and real-time perceived environmental information. While this method can simulate diverse traffic scenarios, the test scenarios are often based on idealized assumptions or artificially constructed, significantly differing from the fixed trajectories of vehicles in real accidents, influenced by environmental factors, vehicle performance, and driver behavior. Furthermore, directly applying the fixed trajectories of real accidents to simulation testing would cause the test vehicle to lose its dynamic decision-making ability; conversely, completely deviating from real-world trajectory planning fails to fully utilize the valuable information from the accident scenario. Existing path planning methods in simulation testing have limitations, failing to fully reproduce the dynamic characteristics of real accident scenarios, thus restricting the accuracy of test results and their practical application value. Summary of the Invention

[0004] This invention provides a virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario, which solves the problem of inaccurate test results caused by scenario distortion in autonomous driving virtual simulation testing.

[0005] The basic solution provided by this invention is a virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario, which specifically includes the following steps: S1: Collection and preprocessing of multi-source data from real accident scenarios, analysis of data to generate point cloud information and visual information of real accident scenes, and extraction of vehicle parameter information, vehicle behavior information, environmental information, and abnormal behavior information of traffic participants, wherein the vehicle behavior information includes fixed trajectories; S2: Construct a virtual simulation scene based on the preprocessed data. Specifically, construct a virtual 3D scene model based on point cloud data and visual information, and set the parameters of the virtual 3D scene model based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules. S3: Virtual simulation test is conducted using a fixed trajectory as a reference trajectory. During the test, the path planning algorithm in the autonomous driving system is used to dynamically adjust the path. S4: Analyze the test data and conduct a multi-dimensional evaluation of the autonomous driving system, generate an analysis report, and generate optimization suggestions based on the analysis report.

[0006] Preferably, S1 includes S1-1) Multi-source data collection of real accident scenarios. The multi-source data of real accident scenarios includes vehicle-mounted equipment data, roadside equipment data, and auxiliary data. The vehicle-mounted equipment data includes vehicle driving data, vehicle-mounted camera video data, and vehicle communication data. The roadside equipment data includes road monitoring data, roadside lidar point cloud data, and radar trajectory data. The auxiliary data includes traffic police on-site survey reports, driver operation data, and data from other road users' mobile devices. S1-2) Perform data cleaning, spatiotemporal alignment, and information fusion on the collected data; S1-3) Generate point cloud information and visual information based on vehicle-mounted equipment data and roadside equipment data; generate vehicle parameter information based on vehicle-mounted equipment data and auxiliary data; and generate vehicle behavior information, environmental information, and abnormal behavior information of traffic participants based on vehicle-mounted equipment data, roadside equipment data, and auxiliary data.

[0007] Preferably, S2 includes S2-1) The point cloud information is analyzed, and the basic framework of the virtual three-dimensional scene model is generated using three-dimensional reconstruction technology. Key features are extracted, and the details of the three-dimensional scene model are filled in using the subdivision surface modeling method. S2-2) Optimize virtual 3D scene models based on visual information through motion recovery structure and texture mapping technology; S2-3) Based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules, set virtual three-dimensional scene model parameters. The virtual three-dimensional scene model parameters include basic scene parameters and core constraint parameters. The basic scene parameters include environmental parameters and basic vehicle attributes. The core constraint parameters include traffic rule constraints, trajectory constraints, and abnormal behavior event triggering.

[0008] More preferably, the environmental parameters include weather type, road surface friction coefficient, and light intensity; the vehicle basic attributes include vehicle type, braking performance, and vehicle size; the traffic rule constraints include speed limits for specific road sections, lane-changing prohibited areas, and following distance thresholds; the trajectory constraints include reference trajectory anchoring, trajectory deviation thresholds, and mandatory constraints on key trajectory nodes; and the abnormal behavior event triggering includes triggering object and behavior definition, triggering timing and condition association, and behavior parameters matching real-world scenarios.

[0009] Preferably, the autonomous driving system includes a perception layer, a decision-making layer, and a control layer. The perception layer is used to generate virtual sensor data, real-time location and speed information of obstacles, and vehicle positioning in the virtual scene; The decision-making layer is used to run path planning algorithms, handle abnormal behavior events of reverse traffic participants, and predict risks. The control layer translates the decisions of the decision-making layer into control commands.

[0010] More preferably, in S3, a dynamic path is generated using a path planning algorithm based on real-time collected virtual environment information, virtual traffic participant behavior change information, virtual vehicle real-time status information, virtual sensor data, simulation system health status information, and combined with traffic rule constraints, reference trajectory parameters, virtual maps, and historical behavior memories.

[0011] In another preferred embodiment, the path planning algorithm includes the spatiotemporal semantic corridor algorithm, the improved TEB algorithm, and the Monte Carlo tree search and deep learning fusion algorithm. During testing, the scene type is first determined, and the corresponding path planning algorithm is selected according to the scene type. The path planning algorithm selection strategy includes: if the scene type is a structured road scene, the spatiotemporal semantic corridor algorithm is used; if the scene type is an unstructured emergency obstacle avoidance scene, the improved TEB algorithm is used; if the scene type is a high uncertainty scene, the Monte Carlo tree search and deep learning fusion algorithm is used.

[0012] Preferably, S4 includes S4-1) Obtain multi-dimensional test data, preprocess the data, and annotate key times; S4-2) The test data is evaluated from multiple dimensions according to the preset quantitative indicators, and the difference between the dynamic path results and the original accident trajectory is analyzed. S4-3) Construct time series causal relationships from the processed test data and analysis results, locate problems through decision tree analysis, and generate optimization suggestions based on the problems; S4-4) Adjust the autonomous driving system and virtual 3D scene model according to the optimization suggestions.

[0013] More preferably, in S4, the quantitative indicators include security, compliance, smoothness, real-time performance, and scenario coverage.

[0014] The principles and advantages of this invention are as follows: 1. A virtual simulation scenario is constructed based on multi-source data from real accident scenarios, and the fixed trajectory of the real accident scenario is used as a reference trajectory for testing. Combining real accident scenarios with virtual simulation testing not only effectively utilizes the multi-dimensional data of real road traffic accident scenarios and simulates real accident scenarios to the greatest extent, but also enables the autonomous driving system to learn from the experience of real accidents in virtual simulation testing, giving full play to the dynamic decision-making ability of the algorithm, thereby improving the accuracy of the test.

[0015] 2. The virtual simulation scenario utilizes a hybrid path planning architecture of "reference trajectory guidance + dynamic decision optimization," which not only leverages the valuable information from real accident trajectories but also empowers the test vehicle with the ability to make dynamic decisions in a virtual environment. This solves the problem in existing technologies where real accident trajectories cannot be effectively used for dynamic testing.

[0016] 3. In virtual simulation scenarios, the path planning algorithm is adaptively switched according to the scenario type, enabling the autonomous driving system to automatically select the most suitable path planning algorithm based on real-time scenario characteristics during virtual simulation testing, thereby improving the accuracy and efficiency of path planning and enhancing the system's adaptability in complex scenarios.

[0017] 4. Establish a closed-loop process from test data collection, analysis and evaluation to algorithm optimization. Through continuous iteration and improvement, continuously enhance the performance of autonomous driving systems in virtual simulation tests based on real accident scenarios, and provide reliable assurance for the practical application of autonomous driving technology. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is a flowchart illustrating the test data analysis and optimization process of this invention. Detailed Implementation

[0019] The following detailed description illustrates the specific implementation method: The specific implementation process is as follows: (See details) Figures 1 to 2 A virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario, specifically including the following steps: S1: Collection and preprocessing of multi-source data from real accident scenarios, analysis of data to generate point cloud information and visual information of real accident scenes, and extraction of vehicle parameter information, vehicle behavior information, environmental information, and abnormal behavior information of traffic participants. S1 includes S1-1) Multi-source data collection for real-world accident scenarios. This multi-source data includes data from vehicle-mounted devices, roadside equipment, and auxiliary data. Vehicle-mounted device data includes vehicle driving data, vehicle camera video data, and vehicle communication data. Roadside equipment data includes road monitoring data, roadside LiDAR point cloud data, and radar trajectory data. Auxiliary data includes traffic police on-site investigation reports, driver operation data, and data from other road users' mobile devices. Specifically, auxiliary data is collected through mobile devices of road users and textual data such as accident reports, traffic police records, and eyewitness testimonies.

[0020] S1-2) Perform data cleaning, spatiotemporal alignment, and information fusion on the collected data; S1-3) Generate point cloud information and visual information based on vehicle-mounted equipment data and roadside equipment data; generate vehicle parameter information based on vehicle-mounted equipment data and auxiliary data; and generate vehicle behavior information, environmental information, and abnormal behavior information of traffic participants based on vehicle-mounted equipment data, roadside equipment data, and auxiliary data.

[0021] Specifically, point cloud information is generated based on roadside lidar point cloud data; visual information is generated based on vehicle-mounted camera video data and road monitoring data; vehicle parameter information is generated based on vehicle driving data and vehicle communication data; vehicle behavior information is generated based on vehicle driving data, vehicle communication data, road monitoring data, roadside lidar point cloud data, and radar trajectory data; environmental information is generated based on vehicle-mounted camera video data, road monitoring data, and roadside lidar point cloud data; and abnormal behavior information of traffic participants is generated based on vehicle communication data, roadside lidar point cloud data, and data from other traffic participants' mobile devices.

[0022] The vehicle parameter information includes vehicle type, dynamic parameters, tire characteristics, transmission performance, etc.; vehicle behavior information includes fixed trajectory, operation intention, emergency behavior, interactive behavior, as well as position, speed, acceleration, etc. at different times; environmental information includes road geometry, traffic facilities, lighting and weather, road surface condition, etc.; abnormal behavior information of traffic participants includes abnormal trajectory, abnormal speed, rule violation, etc.

[0023] The fixed trajectory is specifically obtained by processing lidar point cloud data, vehicle GPS trajectory, and roadside camera video using a Kalman filter-based SLAM optimization algorithm.

[0024] S2: Construct a virtual simulation scene based on the preprocessed data. Specifically, construct a virtual 3D scene model based on point cloud data and visual information, and set the parameters of the virtual 3D scene model based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules. S2 includes S2-1) The point cloud information is analyzed, and the basic framework of the virtual three-dimensional scene model is generated using three-dimensional reconstruction technology. Key features are extracted, and the details of the three-dimensional scene model are filled in using the subdivision surface modeling method. Specifically, a statistical outlier removal algorithm is used to denoise the roadside lidar point cloud data, a Poisson reconstruction algorithm is used to generate a three-dimensional mesh model, and road geometric features such as lane curvature, slope, and buildings are extracted. A subdivision surface modeling method is used to shape the shape details of complex objects. S2-2) Based on visual information, optimize the virtual 3D scene model through structure-of-motion (SFM) and texture mapping technology; specifically, based on visual information such as video data from vehicle-mounted / roadside cameras, reconstruct the 3D skeleton of the virtual scene using SFM technology; use semantic segmentation to identify elements such as traffic signs and traffic lights in the visual information and label them with semantic tags; project multi-view video frames onto the 3D model to generate PBR (physically based rendering) materials, thereby optimizing the virtual 3D scene model; S2-3) Based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules, set virtual three-dimensional scene model parameters. The virtual three-dimensional scene model parameters include basic scene parameters and core constraint parameters. The basic scene parameters include environmental parameters and basic vehicle attributes. The core constraint parameters include traffic rule constraints, trajectory constraints, and abnormal behavior event triggering.

[0025] The environmental parameters include weather type, road surface friction coefficient, and light intensity; the vehicle basic attributes include vehicle type, braking performance, and vehicle size; the traffic rule constraints include speed limits for specific road sections, lane-changing prohibited areas, and following distance thresholds; the trajectory constraints include reference trajectory anchoring, trajectory deviation thresholds, and mandatory constraints on key trajectory nodes; and the abnormal behavior event triggering includes triggering objects and behavior definitions, triggering timing and condition association, and matching behavior parameters with real-world scenarios.

[0026] The autonomous driving system includes a perception layer, a decision-making layer, and a control layer. The perception layer is used to generate virtual sensor data, real-time location and speed information of obstacles, and vehicle positioning in the virtual scene; The decision-making layer is used to run path planning algorithms, handle abnormal behavior events of reverse traffic participants, and predict risks. The control layer, including trajectory tracking controller, actuator model, and vehicle dynamics simulation, is used to convert the decisions of the decision layer into control commands. It is also used to simulate the execution delay of real vehicles and to provide feedback on the actual motion state of the vehicle for closed-loop verification. The control commands include converting the planned path into steering / acceleration / braking commands.

[0027] S3: Virtual simulation test is conducted using a fixed trajectory as a reference trajectory. During the test, the path planning algorithm in the autonomous driving system is used to dynamically adjust the path. In S3, dynamic paths are generated using path planning algorithms based on real-time collected virtual environment information, virtual traffic participant behavior change information, virtual vehicle real-time status information, virtual sensor data, simulation system health status information, and combined with traffic rule constraints, reference trajectory parameters, virtual maps, and historical behavior memories.

[0028] The path planning algorithms include the spatiotemporal semantic corridor algorithm, the improved TEB algorithm, and the Monte Carlo tree search and deep learning fusion algorithm. During testing, the scene type is first determined, and the corresponding path planning algorithm is selected according to the scene type. The path planning algorithm selection strategy includes: if the scene type is a structured road scene, the spatiotemporal semantic corridor algorithm is used; if the scene type is an unstructured emergency obstacle avoidance scene, the improved TEB algorithm is used; if the scene type is a high uncertainty scene, the Monte Carlo tree search and deep learning fusion algorithm is used.

[0029] High uncertainty is defined in ISO 22737:2021 as: complex traffic scenarios with multiple unpredictable interaction objects and environmental perception confidence of less than 70%; SAE J3016 further clarifies: driving environments that simultaneously meet the following three conditions: perception sensor signal-to-noise ratio <15dB; dynamic obstacle behavior prediction accuracy <60%; environmental parameters (such as lighting / weather) exhibit rapid time-varying characteristics; the definition in the China Intelligent Connected Vehicle Standard System (2023 Edition) refers to test scenarios with the following characteristics: 1) multi-source sensor data conflict rate >30%; 2) more than 3 traffic participants with ambiguous behavioral intentions; 3) map positioning error consistently >1.5m.

[0030] S4: Analyze the test data and conduct a multi-dimensional evaluation of the autonomous driving system, generate an analysis report, and generate optimization suggestions based on the analysis report.

[0031] S4 includes S4-1) Obtain multi-dimensional test data, preprocess the data, and annotate key times; The test data includes vehicle status, environmental interaction data, and algorithm intermediate variables. The vehicle status includes position, speed, and acceleration. The environmental interaction data includes obstacle distance and traffic light status. The algorithm intermediate variables include planned path curvature and decision confidence. The test data is checked for time consistency to remove outliers caused by sensor simulation noise. Automatically mark collision moments, rule violation points, emergency intervention trigger moments, etc., and conduct expert confirmation for disputed events; S4-2) The test data is evaluated from multiple dimensions according to the preset quantitative indicators, and the difference between the dynamic path results and the original accident trajectory is analyzed. Quantitative indicators include security, compliance, smoothness, real-time performance, and scenario coverage; The safety features include the probability of collision risk and the success rate of obstacle avoidance. Collision risk probability = (Number of tests where a collision occurred / Total number of tests) × 100%; Obstacle avoidance success rate = (Number of successful obstacle avoidance tests / Total number of obstacle avoidance tests) × 100%; The compliance includes the number of traffic rule violations and the average following distance; The smoothness includes the rate of change of acceleration and the rate of change of path curvature; The real-time performance includes the time consumed in a single planning operation; The scenario coverage includes the number of scenario types that can be effectively addressed.

[0032] S4-3) Construct time series causal relationships from the processed test data and analysis results, locate problems through decision tree analysis, and generate optimization suggestions based on the problems; S4-4) Adjust the autonomous driving system and virtual 3D scene model according to the optimization suggestions.

[0033] In S4, the optimization suggestions include suggestions for adjusting the autonomous driving system and suggestions for adjusting the virtual 3D scene model.

[0034] A. Optimization suggestions for adjusting the autonomous driving system could be: If the problem is low perception confidence and high data conflict rate from multiple sensors... Improvements can be made 1) Improve fusion algorithms, such as applying deep learning-based fusion networks or probabilistic grid fusion, to improve the stability and accuracy of target tracking.

[0035] 2) Enhance anti-interference capabilities. For low light, rain, snow and fog weather, add corresponding sensor noise models (such as camera glare, lidar point cloud attenuation, radar false alarm) in the simulation, and train the data augmentation and denoising capabilities of the perception model accordingly.

[0036] 3) Adjust sensor parameters, such as the sensor's installation position, viewing angle, and resolution, to optimize FOV (field of view) coverage and reduce blind spots.

[0037] If the accuracy of behavioral prediction is low, decision-making will be hesitant or rule-breaking will occur. Improvements can be made 1) Adopt more refined decision-making rules, adding more factors and state branches to the decision tree or state machine. For example, consider not only the distance to the vehicle in front, but also the speed of vehicles approaching from the side and rear, the remaining time of the traffic light, etc., to make decision parameters such as following distance threshold and lane changing timing more adaptive and safer.

[0038] 2) Introduce reinforcement learning and use a large amount of "accident" or "risk" data generated in the simulation environment as negative samples to train the reinforcement learning model, so that it can learn to make safer and more efficient decisions in complex interactions.

[0039] 3) Optimize the risk prediction model. Based on historical test data, recalibrate the parameters of the risk prediction module to make it predict potential collisions earlier and more accurately.

[0040] If the problem is a large path tracking error, the control commands are not smooth (high abruptness). Optimization suggestions can be 1) Optimize controller parameters. For trajectory tracking controllers such as PID or MPC, adjust their weight matrix. For example, increase the weight of tracking error to improve accuracy, or increase the weight of control rate of change to improve comfort.

[0041] 2) Calibrate the actuator model. Based on real vehicle data, more accurately simulate the delay, response speed and limits (such as maximum steering angular velocity, maximum acceleration / deceleration) of the actuator model, making the control simulation closer to reality.

[0042] 3) Optimize the pre-aiming strategy by increasing the controller's pre-aiming distance or adopting a smarter adaptive pre-aiming algorithm to enable it to react smoothly to curves or obstacles ahead earlier.

[0043] B. Optimization suggestions for adjusting the virtual 3D scene model If the problem is insufficient scene coverage, the system cannot handle certain edge scenarios. Optimization suggestions can be: 1) Generate edge cases: Based on the original accident data, create more variations of high-uncertainty scenarios by using adversarial generation or parameter perturbation methods. For example, make pedestrians rush out at more unusual angles, or make multiple vehicles behave abnormally at the same time.

[0044] 2) Expand the scenario library by adding the weak scenarios found in this test (such as unlit intersections at night under specific weather conditions) to the regular regression test scenario library.

[0045] If the problem is that the model parameters deviate from the real world, causing the test results to be distorted. Optimization suggestions can be: 1) Calibrate physical parameters. Based on actual measurements or more authoritative data sources, adjust the road friction coefficient, vehicle dynamic parameters (such as air drag coefficient and tire lateral stiffness) to make the vehicle dynamics simulation more accurate.

[0046] 2) Refine the behavior model and optimize the behavioral parameters of traffic participants (especially those exhibiting abnormal behavior). For example, make the speed distribution of pedestrians crossing the road, the cutting angle and acceleration of vehicles cutting in more consistent with the distribution obtained from statistical data.

[0047] If the problem stems from incomplete or inaccurate traffic rules, making it impossible to effectively detect violations. Optimization suggestions can be: 1) Supplement local traffic regulations by adding more localized traffic rule details to the traffic rule constraints, such as tidal flow lanes, waiting areas, and yielding rules.

[0048] 2) Dynamically adjust constraints so that rule constraints are no longer fixed values, but can change dynamically according to environmental parameters. For example, the safe following distance threshold should automatically increase with the degree of road slipperiness (low friction coefficient) and visibility (heavy rain).

[0049] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario, characterized in that, Specifically, the following steps are included: S1: Collection and preprocessing of multi-source data from real accident scenarios, analysis of data to generate point cloud information and visual information of real accident scenes, and extraction of vehicle parameter information, vehicle behavior information, environmental information, and abnormal behavior information of traffic participants, wherein the vehicle behavior information includes fixed trajectories; S2: Construct a virtual simulation scene based on the preprocessed data. Specifically, construct a virtual 3D scene model based on point cloud data and visual information, and set the parameters of the virtual 3D scene model based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules. S3: Virtual simulation test is conducted using a fixed trajectory as a reference trajectory. During the test, the path planning algorithm in the autonomous driving system is used to dynamically adjust the path. S4: Analyze the test data and conduct a multi-dimensional evaluation of the autonomous driving system, generate an analysis report, generate optimization suggestions based on the analysis report, and perform optimization.

2. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 1, characterized in that: S1 includes S1-1) Multi-source data collection of real accident scenarios. The multi-source data of real accident scenarios includes vehicle-mounted equipment data, roadside equipment data, and auxiliary data. The vehicle-mounted equipment data includes vehicle driving data, vehicle-mounted camera video data, and vehicle communication data. The roadside equipment data includes road monitoring data, roadside lidar point cloud data, and radar trajectory data. The auxiliary data includes traffic police on-site survey reports, driver operation data, and data from other road users' mobile devices. S1-2) Perform data cleaning, spatiotemporal alignment, and information fusion on the collected data; S1-3) Generate point cloud information and visual information based on vehicle-mounted equipment data and roadside equipment data; generate vehicle parameter information based on vehicle-mounted equipment data and auxiliary data; and generate vehicle behavior information, environmental information, and abnormal behavior information of traffic participants based on vehicle-mounted equipment data, roadside equipment data, and auxiliary data.

3. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 1, characterized in that: S2 includes S2-1) The point cloud information is analyzed, the basic framework of the virtual three-dimensional scene model is generated using three-dimensional reconstruction technology, key features are extracted, and the details of the three-dimensional scene model are filled in using the subdivision surface modeling method. S2-2) Optimize virtual 3D scene models based on visual information through motion recovery structure and texture mapping technology; S2-3) Based on vehicle parameter information, vehicle behavior information, environmental information, abnormal behavior information of traffic participants, and traffic rules, set virtual three-dimensional scene model parameters. The virtual three-dimensional scene model parameters include basic scene parameters and core constraint parameters. The basic scene parameters include environmental parameters and basic vehicle attributes. The core constraint parameters include traffic rule constraints, trajectory constraints, and abnormal behavior event triggering.

4. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 3, characterized in that: The environmental parameters include weather type, road surface friction coefficient, and light intensity; the vehicle basic attributes include vehicle type, braking performance, and vehicle size; the traffic rule constraints include speed limits for specific road sections, lane-changing prohibited areas, and following distance thresholds; the trajectory constraints include reference trajectory anchoring, trajectory deviation thresholds, and mandatory constraints on key trajectory nodes; and the abnormal behavior event triggering includes triggering objects and behavior definitions, triggering timing and condition association, and matching behavior parameters with real-world scenarios.

5. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 1, characterized in that: The autonomous driving system includes a perception layer, a decision-making layer, and a control layer. The perception layer is used to generate virtual sensor data, real-time location and speed information of obstacles, and vehicle positioning in the virtual scene; The decision-making layer is used to run path planning algorithms, handle abnormal behavior events of reverse traffic participants, and predict risks. The control layer translates the decisions of the decision-making layer into control commands.

6. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 5, characterized in that: In S3, dynamic paths are generated using path planning algorithms based on real-time collected virtual environment information, virtual traffic participant behavior change information, virtual vehicle real-time status information, virtual sensor data, simulation system health status information, and combined with traffic rule constraints, reference trajectory parameters, virtual maps, and historical behavior memories.

7. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 6, characterized in that: The path planning algorithms include the spatiotemporal semantic corridor algorithm, the improved TEB algorithm, and the Monte Carlo tree search and deep learning fusion algorithm. During testing, the scene type is first determined, and the corresponding path planning algorithm is selected according to the scene type. The path planning algorithm selection strategy includes: if the scene type is a structured road scene, then the spatiotemporal semantic corridor algorithm is adopted; If the scenario type is an unstructured emergency obstacle avoidance scenario, the improved TEB algorithm is used; if the scenario type is a high uncertainty scenario, a Monte Carlo tree search and deep learning fusion algorithm is used.

8. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 1, characterized in that: S4 includes S4-1) Obtain multi-dimensional test data, preprocess the data, and annotate key times; S4-2) The test data is evaluated from multiple dimensions according to the preset quantitative indicators, and the difference between the dynamic path results and the original accident trajectory is analyzed. S4-3) Construct time series causal relationships from the processed test data and analysis results, locate problems through decision tree analysis, and generate optimization suggestions based on the problems; S4-4) Adjust the autonomous driving system and virtual 3D scene model according to the optimization suggestions.

9. The virtual simulation dynamic path testing method based on a fixed trajectory in an accident scenario according to claim 8, characterized in that: In S4, quantitative metrics include security, compliance, smoothness, real-time performance, and scenario coverage.

Images