CN-121681254-B - Expressway scene automatic construction method and system based on actual test data

CN121681254BCN 121681254 BCN121681254 BCN 121681254BCN-121681254-B

Abstract

The invention provides an automatic expressway scene construction method and system based on actual test data, and relates to the technical field of automatic driving. The method comprises the steps of classifying and synchronously processing actual road test data to obtain multi-source information, preprocessing and extracting features to obtain static, dynamic and risk association features, mining core influence factors and determining risk mutation thresholds of the core influence factors to split scenes into minimum functional units, performing time sequence analysis on the dynamic features to form scene dynamic evolution rules, constructing a risk assessment model, calculating risk scores and classifying the risk scores to generate high-risk scenes aiming at the high-risk classes, establishing an association map to generate complete test scenes, constructing a digital twin body to continuously detect actual road network and traffic flow changes, dynamically updating the minimum functional units and correcting scene library parameters, and realizing automatic, efficient construction and continuous updating of the dynamic high-fidelity test scenes from the real data.

Inventors

WANG GUANGYU
SONG SHIPING
ZHANG CHENG
ZHANG YITENG
QU GE
YU XIAOJUN

Assignees

中汽智能科技(天津)有限公司

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (10)

1. The automatic expressway scene construction method based on the actual test data is characterized by comprising the following steps of: Classifying and synchronizing the collected actual road test data to obtain environment perception data, road network data, main vehicle information, traffic participant information and barrier information; preprocessing and extracting features of environment perception data, road network data, host vehicle information, traffic participant information and barrier information to obtain static features, dynamic features and risk association features; Carrying out statistical analysis on the static features, the dynamic features and the risk association features to mine core influence factors, determining risk mutation threshold values of the core influence factors, and splitting a scene into minimum functional units based on the core influence factors and corresponding threshold value intervals; performing time sequence analysis on the dynamic characteristics to refine scene evolution logic, and integrating the evolution logic to form scene dynamic evolution rules; Based on the core influence factors, the risk mutation threshold values of the core influence factors and the minimum functional units, weight is distributed to each core influence factor to construct a risk assessment model, comprehensive risk scores of the minimum functional units are calculated by using the risk assessment model to divide risk grades, and parameter combination and physical constraint verification are carried out on the core influence factors corresponding to the high risk grades to generate a high risk scene; Establishing a correlation map between a preset test requirement, a scene dynamic evolution rule and a high-risk scene, screening the matched scene dynamic evolution rule and the high-risk scene based on the correlation map, combining the scene dynamic evolution rule and the high-risk scene to generate a complete test scene, and outputting scene parameters; based on continuously collected actual road test data, constructing and updating a highway digital twin body, updating a minimum functional unit according to road network or traffic flow change characteristics detected by the digital twin body, and carrying out deviation calculation and parameter correction on scenes in a scene library based on the latest actual road test data and complete test scenes.
2. The automatic expressway scene construction method based on actual test data, as set forth in claim 1, wherein the environment-aware data includes weather, visibility, and road surface conditions, the road network data includes the number of lanes, curvature, and gradient, the host vehicle information includes vehicle basic information and host vehicle state information, the host vehicle state information is time-series data, the traffic participant information includes participation vehicle basic information and movement state information, the movement state information is time-series data, and the obstacle information includes obstacle properties, space information, and time-series information.
3. The automatic expressway scene construction method according to claim 1, wherein preprocessing and feature extraction are performed on environment-aware data, road network data, host vehicle information, traffic participant information, and obstacle information to obtain static features, dynamic features, and risk-related features, comprising: Preprocessing environment perception data, road network data, host vehicle information, traffic participant information and barrier information, cleaning invalid data and unifying data formats; And extracting features of the preprocessed environment perception data, road network data, host vehicle information, traffic participant information and barrier information, extracting road network topology, infrastructure attributes and lane attributes as static features, extracting time-series vehicle behaviors, environment parameter evolution sequences and traffic participant behavior sequences as dynamic features, extracting coupling features of vehicle parameters and environment parameters as risk association features, and obtaining the static features, the dynamic features and the risk association features.
4. The automatic expressway scene construction method according to claim 1, wherein the statistical analysis is performed on static features, dynamic features, and risk association features to mine core influence factors, and a risk mutation threshold of each core influence factor is determined, and the scene is split into minimum functional units based on the core influence factors and their corresponding threshold intervals, comprising: carrying out statistical analysis on the static characteristics, the dynamic characteristics and the risk association characteristics, and mining out core influence factors; Based on actual measurement accident data of the core influence factors, calculating risk probability of each core influence factor under different values, and determining a risk mutation threshold of each core influence factor according to preset risk mutation conditions; Dividing threshold intervals of the core influence factors based on risk mutation thresholds of the core influence factors; and combining the threshold intervals of the core influence factors to generate minimum functional units covering different risk dimensions, wherein each minimum functional unit corresponds to a group of core influence factors and unique combinations of the threshold intervals thereof.
5. The automatic expressway scene building method according to claim 1, wherein determining a threshold interval corresponding to the core influence factor includes: Calculating a multi-factor coupling risk mutation coefficient based on the risk mutation coefficient of each core influence factor; and determining the value combination of each core influence factor which enables the multi-factor coupling risk mutation coefficient to meet the preset coupling risk lifting condition as a threshold value interval.
6. The automatic expressway scene construction method based on actual test data, as set forth in claim 1, wherein performing time-series analysis on the dynamic characteristics to refine scene evolution logic, integrating the evolution logic to form scene dynamic evolution rules, comprises: Analyzing the time sequence of the dynamic characteristics, and analyzing the change rule of scene elements along with time; Based on a change rule, determining a triggering condition for starting scene evolution, wherein the triggering condition comprises vehicle state change, environment parameter change or a road surface abnormal event; Based on a change rule, constructing an evolution path of scene elements changing along with time, wherein the evolution path comprises a vehicle behavior evolution chain, an environment parameter evolution chain or a traffic participant interaction evolution chain; defining a termination condition for ending scene evolution based on a change rule, wherein the termination condition comprises that a vehicle recovers a stable state or a risk source disappears; And integrating the triggering condition, the evolution path and the termination condition to form a scene dynamic evolution rule.
7. The automatic expressway scenario construction method according to claim 1, wherein, based on the core influence factors and the risk mutation thresholds thereof and the minimum functional units, weights are assigned to each core influence factor to construct a risk assessment model, the risk assessment model is used to calculate the comprehensive risk score of each minimum functional unit to divide the risk level, and parameter combination and physical constraint verification are performed for the core influence factors corresponding to the high risk level, so as to generate a high risk scenario, comprising: Assigning a weight to each core impact factor based on the risk mutation coefficient of each core impact factor; based on the risk mutation coefficient and the corresponding weight of each core influence factor, calculating the comprehensive risk score of each minimum functional unit; dividing the minimum functional unit into different risk grades based on a preset scoring interval and a comprehensive risk score; aiming at the core influence factors corresponding to the high risk level, dividing parameter levels in a threshold interval of the core influence factors to generate a parameter level combination; And carrying out physical constraint verification on the parameter level combination, and eliminating the combination which does not accord with the physical rule to obtain a verified high-risk scene.
8. The automatic expressway scene construction method according to claim 1, wherein establishing a preset association map between the test requirements and the scene dynamic evolution rules and high risk scenes, screening the matched scene dynamic evolution rules and high risk scenes based on the association map, combining to generate a complete test scene and outputting scene parameters, comprises: Combing the preset test requirements of the automatic driving system, analyzing the relevance between the preset test requirements and the scene dynamic evolution rules and between the preset test requirements and the high-risk scenes, and establishing a relevance map; Receiving a test requirement input by a user, and screening out a scene dynamic evolution rule and a high-risk scene matched with the test requirement input by the user based on the association graph; And combining the matched high-risk scenes according to scene logic based on the screened scene dynamic evolution rules to generate a complete test scene, and outputting scene parameters comprising duration and risk point distribution.
9. The automatic expressway scene construction method according to claim 1, wherein the expressway digital twin is constructed and updated based on continuously collected actual road test data, and the minimum functional unit is updated based on road network or traffic flow variation characteristics detected by the digital twin, and the scene in the scene library is subjected to deviation calculation and parameter correction based on the latest actual road test data and complete test scenes, comprising: constructing a digital twin body of the expressway based on continuously collected actual road test data, so that the digital twin body synchronizes the road network topology, traffic flow and environmental state of the expressway; when the digital twin body detects the road network or traffic flow change, the change characteristics are automatically extracted; based on the change characteristics, updating the minimum functional unit, and newly adding or modifying the minimum functional unit corresponding to the change characteristics; calculating the deviation of the physical parameters of the scene in the scene library and the actual measured physical parameters based on the latest actual road test data and the complete test scene at regular intervals; When the deviation exceeds a preset threshold, correcting physical parameters of the corresponding scene to keep the consistency of the scene and the actual state of the expressway.
10. An automatic expressway scene construction system based on actual test data, characterized in that the system adopts the automatic expressway scene construction method based on actual test data as claimed in any one of claims 1 to 9, and the system comprises: The data acquisition and synchronization module is used for performing the steps of classifying and synchronizing the acquired actual road test data to obtain environment sensing data, road network data, host vehicle information, traffic participant information and barrier information; the feature extraction module is used for carrying out preprocessing and feature extraction on environment perception data, road network data, host vehicle information, traffic participant information and barrier information to obtain static features, dynamic features and risk association features; The minimum functional unit splitting module is used for carrying out statistical analysis on the static characteristics, the dynamic characteristics and the risk association characteristics to mine core influence factors, determining risk mutation threshold values of the core influence factors, and splitting a scene into minimum functional units based on the core influence factors and threshold value intervals corresponding to the core influence factors; The scene dynamic evolution rule construction module is used for executing the steps of carrying out time sequence analysis on the dynamic characteristics to refine scene evolution logic and integrating the evolution logic to form scene dynamic evolution rules; The risk quantification and high-risk scene generation module is used for executing the steps of distributing weights for all the core influence factors based on the core influence factors, the risk mutation threshold values and the minimum functional units thereof to construct a risk assessment model, calculating the comprehensive risk score of each minimum functional unit by using the risk assessment model to divide risk grades, and carrying out parameter combination and physical constraint verification on the core influence factors corresponding to the high risk grades to generate a high-risk scene; The intelligent arrangement and output module of the test scene is used for executing the steps of establishing a preset association graph between the test requirement, the scene dynamic evolution rule and the high-risk scene, screening the matched scene dynamic evolution rule and the high-risk scene based on the association graph, combining the scene dynamic evolution rule and the high-risk scene to generate a complete test scene and outputting scene parameters; The scene library dynamic updating and maintaining module is used for executing the steps of constructing and updating a highway digital twin body based on continuously collected actual road test data, updating a minimum functional unit according to road network or traffic flow change characteristics detected by the digital twin body, and carrying out deviation calculation and parameter correction on scenes in the scene library based on the latest actual road test data and complete test scenes.

Description

Expressway scene automatic construction method and system based on actual test data Technical Field The invention relates to the technical field of automatic driving, in particular to an automatic expressway scene construction method and system based on actual test data. Background In the development and testing process of an automatic driving system, the construction of a highway scene is a key link for verifying the functions and the safety of the system. Currently, the construction of test scenes is mostly dependent on manual experience design or simple data playback, and has significant limitations. The artificial design scene is limited in coverage, complexity and diversity of a real road environment are difficult to comprehensively reflect, particularly, the mining of a potential high-risk scene is not enough and deep, while the simple data playback is based on real data, but the scene elements are not deeply analyzed, risk quantized and logically reconstructed, so that the structured complete scene meeting specific test requirements cannot be flexibly combined and generated. In addition, the road environment and the traffic flow state continuously change, the existing method is difficult to update the scene library timely and automatically to keep the consistency with the real world, so that the test scene is gradually disjointed from the real situation, and the effectiveness and reliability of the test are reduced. Therefore, a technical solution capable of automatically and efficiently mining and quantifying risks from actual road data, constructing a logical scene and continuously and dynamically updating is needed. Disclosure of Invention In order to solve the above problems in the prior art, the first aspect of the present invention provides an automatic expressway scene construction method based on actual test data, including: s1, classifying and synchronizing acquired actual road test data to obtain environment perception data, road network data, main vehicle information, traffic participant information and barrier information; s2, preprocessing and extracting features of environment perception data, road network data, host vehicle information, traffic participant information and barrier information to obtain static features, dynamic features and risk association features; S3, carrying out statistical analysis on the static features, the dynamic features and the risk association features to mine core influence factors, determining risk mutation thresholds of the core influence factors, and splitting a scene into minimum functional units based on the core influence factors and corresponding threshold intervals; S4, carrying out time sequence analysis on the dynamic characteristics to refine scene evolution logic, and integrating the evolution logic to form scene dynamic evolution rules; S5, based on the core influence factors, the risk mutation threshold values of the core influence factors and the minimum functional units, weight is distributed to each core influence factor to construct a risk assessment model, the risk assessment model is utilized to calculate the comprehensive risk score of each minimum functional unit to divide the risk grade, and parameter combination and physical constraint verification are carried out on the core influence factors corresponding to the high risk grade to generate a high risk scene; s6, establishing a correlation map between a preset test requirement, a scene dynamic evolution rule and a high-risk scene, screening the matched scene dynamic evolution rule and the high-risk scene based on the correlation map, combining the scene dynamic evolution rule and the high-risk scene to generate a complete test scene, and outputting scene parameters; S7, constructing and updating a highway digital twin body based on continuously collected actual road test data, updating a minimum functional unit according to road network or traffic flow change characteristics detected by the digital twin body, and carrying out deviation calculation and parameter correction on scenes in a scene library based on the latest actual road test data and complete test scenes. Compared with the prior art, the invention has the beneficial effects that: The invention forms a closed loop and automatic construction and updating flow from data to scenes, and solves the problems of insufficient coverage, insufficient risk quantification, poor flexibility and updating lag of scene construction in the background technology. Firstly, classification and synchronous processing are carried out on actual road test data, so that the consistency of multi-source heterogeneous data such as environment, road network, vehicles, obstacles and the like in space-time is ensured, and a high-quality data base is provided for subsequent analysis. Then, by preprocessing and extracting the data, static characteristics, dynamic characteristics and risk association characteristics are systematically obtained, digital ana