Automatic driving test scene library construction method based on real traffic data

Abstract

The invention discloses an automatic driving test scene library construction method based on real traffic data, and relates to the technical field of automatic driving, and the method comprises the following steps: based on a selective sensor fusion framework, dynamically adjusting a fusion strategy of a multi-modal sensor according to a current driving environment, and obtaining corresponding scene elements; performing hierarchical classification on scene elements, constructing a risk assessment model, calculating a comprehensive risk score, and preliminarily dividing risk levels; constructing a rule-based classifier by adopting an association rule mining technology on the basis of results of hierarchical classification and preliminary risk grading, and carrying out risk grading on the scene to be evaluated; and the scene elements and the risk levels are stored in a structured manner, and an automatic driving scene library supporting multi-dimensional query is constructed. According to the method, the characteristics of the traffic scene can be captured more comprehensively, scene elements can be identified more accurately, the risk levels of the scene can be divided scientifically, and the scene library is constructed by combining the scene elements and the risk levels, so that the diversity and pertinence of the test scene are improved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method for constructing an autonomous driving test scenario library based on real traffic data. Background Art

[0002] With the rapid development of autonomous driving technology, its commercialization has made tremendous progress. To accelerate the implementation of autonomous driving technology, it is necessary to build autonomous driving test scenarios to extract a rich and diverse range of test cases to comprehensively test autonomous driving systems. However, existing methods for building test scenario libraries have some shortcomings. Existing scenario library construction methods often rely on manual labeling with low automation levels, and risk assessment indicators are vague and difficult to quantify. There is a lack of detailed classification of scenario elements and risk types, which limits the risk assessment and management capabilities of test scenario libraries and makes it difficult to effectively support the diverse testing needs of autonomous driving systems. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for constructing an autonomous driving test scenario library based on real traffic data. By comprehensively capturing the characteristics of traffic scenes, identifying scene elements and scientifically dividing the risk levels of scenes, a scenario library is constructed, thereby improving the diversity and pertinence of test scenarios.

[0004] To achieve the above objectives, the present invention provides a method for constructing an autonomous driving test scenario library based on real traffic data, comprising the following steps:

[0005] Step 1. Based on the selective sensor fusion framework, dynamically adjust the fusion strategy of the multimodal sensor according to the current driving environment. The selective sensor fusion framework is configured as follows:

[0006] There are multiple convolutional neural networks (CNNs) set up, and each sensor data input is sent to its corresponding convolutional neural network (CNN) to extract the data features F i , according to the data feature F i , identifying the context information of the current scene, and selecting a sensor fusion strategy suitable for the current scene based on the context information; using a preset deep learning model branch to process the corresponding sensor fusion strategy to obtain a target scene result and corresponding scene elements;

[0007] Step 2. The scenario elements obtained in step 1 are standardized by unifying their names and data formats, and then categorized into different levels. Key elements in each level of the scenario are selected as risk assessment indicators. A risk assessment model is constructed to calculate a comprehensive risk score and preliminarily classify the risk levels.

[0008] Step 3. Based on the hierarchical classification results of the scenario elements and the preliminary risk level classification results in Step 2, association rule mining technology is used to mine the association rules between the scenario elements and risk levels, and a rule-based classifier is constructed to classify the risk level of the scenario to be assessed;

[0009] Step 4. Structured storage of scenario elements and risk levels, building an autonomous driving scenario library that supports multi-dimensional queries, and the ability to flexibly filter out test sets with specific elements and risk levels based on different testing requirements.

[0010] There are multiple deep learning model branches, each of which processes a specific sensor fusion strategy;

[0011] The processing result of the deep learning model branch on the sensor fusion strategy includes at least a bounding box, a confidence level, and a type label of the target.

[0012] It also includes performing non-maximum suppression or weighted box fusion on the processing result of the deep learning model branch on the sensor fusion strategy.

[0013] Step 2: The scene elements are divided into six layers of scene elements:

[0014] The main vehicle layer includes the status of the vehicle under test;

[0015] The weather layer includes external environmental conditions that affect perception and vehicle dynamics;

[0016] Traffic participant layer, including dynamic obstacle types and behavior intentions;

[0017] Temporary construction layer, including non-standardized road change elements;

[0018] Traffic facilities layer, including fixed traffic management and guidance facilities;

[0019] The road topology layer includes the structural features of the road network.

[0020] The calculation formula for constructing the risk assessment model described in step 2 is:

[0021]

[0022] Among them, R represents the comprehensive risk score, x i is the score of risk assessment index i, w i is the weight of risk assessment indicator i, and n represents the number of risk assessment indicators;

[0023] Scenarios are divided into three risk levels based on the scenario risk percentage range, namely normal, critical and hazardous scenarios. The risk levels are divided by percentage: normal (0% ≤ R < 30%), critical (30% ≤ R < 70%), and hazardous (R ≥ 70%).

[0024] Scenario risk percentage R ％ The calculation formula is as follows:

[0025]

[0026] where R min Indicates the minimum risk score, R max Indicates the highest risk score.

[0027] In step 3, the association rule mining technology is the Apriori algorithm;

[0028] Through the layer-by-layer iteration mechanism of the Apriori algorithm, the association between a single factor and multiple factor combinations on risk levels is discovered, forming an original rule base of factor combination-risk level;

[0029] The generated original rule base is cleaned and deduplicated, and weights are assigned to all cleaned rules to obtain a rule-based classifier. The scene elements of each scenario to be evaluated are matched with the rules in the classifier. The weight of each successfully matched rule is calculated, and a comprehensive score is calculated for the risk level of the scenario. The calculation formula is:

[0030]

[0031] Where n is the number of matching rules, X i is the score of risk assessment indicator i, Y is the risk level, and an aggressive strategy is used to divide the risk levels.

[0032] Adopting an aggressive strategy to categorize risk levels means:

[0033] The risk level with the highest comprehensive score is taken as the final classification result;

[0034] If the calculated high risk comprehensive score = low risk comprehensive score, the risk is classified as high.

[0035] The multimodal sensor includes at least a camera and a laser radar.

[0036] A computer-readable storage medium stores computer program instructions, which can be executed by a processor to implement the above-mentioned method for constructing an autonomous driving test scenario library based on real traffic data.

[0037] 3) Beneficial effects:

[0038] Compared with the existing technology, this method for constructing an autonomous driving test scenario library based on real traffic data has the following beneficial effects:

[0039] First, the method of the present invention processes multimodal sensor data collected in traffic scenes to identify and extract traffic scene features. It then combines a hierarchical structural analysis method for scene elements to construct a risk assessment model for preliminary risk classification. It then uses an association rule mining algorithm to extract association rules between scene elements and risk levels, and constructs a rule-based scene classifier to automate scene risk classification. Finally, it constructs a library of autonomous driving test scenarios based on scene elements and risk levels. This method can more comprehensively capture traffic scene features, more accurately identify scene elements, and scientifically classify scene risk levels. Combining these two methods to build a scenario library increases the diversity and pertinence of test scenarios, providing stronger support for the development and verification of autonomous driving technology.

[0040] Second, the construction method provided by this invention not only improves the coverage of test scenarios but also enables rapid retrieval of test scenarios with specific characteristics and risk levels based on different testing requirements. This will help improve the safety and stability of autonomous driving systems and accelerate their commercialization. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the sensor data fusion and feature recognition process provided by an embodiment of the present invention;

[0042] Figure 2 This is an overall schematic diagram of the scene library construction method of the present invention;

[0043] Figure 3 This is an overall schematic diagram of the scene library construction method of the present invention; DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] like Figure 1-3 As shown, the present invention provides a technical solution: a method for constructing an autonomous driving test scenario library based on real traffic data, comprising the following steps:

[0046] Step 1. Based on the selective sensor fusion framework, dynamically adjust the fusion strategy of the multimodal sensor according to the current driving environment. The selective sensor fusion framework is configured as follows:

[0047] There are multiple convolutional neural networks (CNNs) set up, and each sensor data input is sent to its corresponding convolutional neural network (CNN) to extract the data features F i , according to the data feature F i , identifying the context information of the current scene, and selecting a sensor fusion strategy suitable for the current scene based on the context information; using a preset deep learning model branch to process the corresponding sensor fusion strategy to obtain a target scene result and corresponding scene elements;

[0048] The autonomous driving system is equipped with a variety of sensors to collect environmental information. Existing sensor fusion technologies mostly use static fusion strategies, which have limitations in complex scenarios. For example, the front fusion is prone to fail when the sensor data is noisy or occluded, and the back fusion cannot fully utilize the multi-sensor features, resulting in a decrease in estimation accuracy. In order to improve data fusion and feature recognition rate, in the specific implementation process, the present invention adopts the HydraFusion selective sensor fusion framework to form a multimodal sensor fusion strategy. It learns to identify the contextual information of the driving scene (such as weather, lighting, road type, etc.), and selects the best sensor combination for fusion based on this information, thereby improving the robustness of perception without increasing computational complexity and energy consumption. Its workflow diagram is as follows: Figure 1 shown.

[0049] HydraFusion accepts input data from multiple sensors, and each sensor data is fed into its own corresponding convolutional neural network (CNN), which is called a "Stem". Stem converts the sensor data into data feature representation for subsequent context recognition and fusion strategy selection. For each sensor S i , its corresponding Stem i The data feature F will be extracted i :

[0050] F i =Stem i (S i )

[0051] i is a sensor (camera, lidar, etc.), and the features of all sensors are combined into a data feature set F:

[0052] F={F radar ,F camera …}

[0053] The features of all sensors are combined into a data feature set F, and the feature set of the lidar is F radar , the camera’s feature set is F camera ;

[0054] The gating module uses the contextual information of the current scene, including features such as light intensity and road type, to identify the context. The contextual information is output by the context recognition model, and the gating module uses this contextual information to dynamically select the optimal sensor fusion strategy for the current scene. Specifically, it determines which sensor data should be fused and how to fuse them based on the current contextual information.

[0055] According to the optimal sensor fusion strategy determined by the gating module, a set of selected deep learning model branches are executed. Each branch model processes a specific sensor combination (the branch can be a single sensor data or a multi-sensor fusion branch) and generates a detection result. Radar branch detection for a scene radar and camera and lidar fusion branch detection radar+camera :

[0056] O radar = radar (F radar )

[0057] O radar+camera = radar+camera (F radar +F camera )

[0058] The output of each branch O * A set of object detection results, including the object's bounding box, confidence score, and type label. Through non-maximum suppression or weighted box fusion (WBF), the best detection result is selected to improve the accuracy and robustness of object detection.

[0059] The expression formula using the WBF algorithm is as follows:

[0060]

[0061] For each scene, there are m branches of detection results. The target frame coordinates detected by the k-th branch are (μ k,1 , v k,1 , μ k,2 , v k,2 ), the confidence level is C k . and Represents the coordinates of the final fused bounding box, and finally obtains the detection result of each target O = Fusion (all O * ).

[0062] The framework outputs a list of detected targets (target category, quantity, etc.) for each scene, along with the identified scene context (weather type, lighting conditions, road type, etc.). This information, which reflects the characteristics, state, and dynamic changes of a scene, is called scene elements. After extracting elements for a single scene, the target categories and context information appearing in all scenes are aggregated and standardized through data cleaning, naming standardization, and data formatting to ensure data accuracy and consistency, forming a macroscopic traffic scene element for subsequent analysis.

[0063] All data of each scene are sorted according to the results of unified standardization to ensure the integrity and consistency of each scene element information, and each scene and its element information are organized into a structured format such as JSON or XML.

[0064] Step 2. The scenario elements obtained in step 1 are standardized by unifying the names and data formats, and then categorized into layers. Key elements in each layer of the scenario are selected as risk assessment indicators. A risk assessment model is constructed to calculate the comprehensive risk score and preliminarily classify the risk levels.

[0065] A six-layer scenario element classification framework is employed to systematically decouple and categorize the macroscopic traffic scenario elements summarized in Step 1 through a structured hierarchical approach. The six layers encompass a comprehensive set of scenario elements at six levels: the primary vehicle layer, the weather layer, the traffic participant layer, the temporary construction layer, the traffic facility layer, and the road topology layer. Each layer independently characterizes risk characteristics at a specific dimension and, through interactions (such as the coupled influence of the weather layer on the perceived visibility of the traffic participant layer), forms the foundation for systematic risk analysis.

[0066] The classification of the six-layer scene elements is as follows:

[0067] Main vehicle layer: covers the status of the vehicle under test, such as speed, yaw angle, etc.

[0068] Weather layer: External environmental conditions that affect perception and vehicle dynamics, such as lighting, precipitation, wind speed, etc.

[0069] Traffic participant layer: dynamic obstacles and their behavioral intentions, such as motor vehicles, pedestrians, etc.

[0070] Temporary construction layer: non-standard temporary road change elements, such as guide gestures, temporary traffic lights, etc.

[0071] Traffic facilities layer: fixed traffic management and guidance facilities, such as traffic light phases, speed limit signs, virtual and real lane lines, etc.

[0072] Road topology layer: structural features of the road network, such as the number of lanes, ramp connection points, intersection types, etc.

[0073] From all scenario factors, several key factors that have a greater impact on risk assessment are selected as evaluation indicators and assigned weight coefficients to the indicators to divide the risk levels.

[0074] Scoring criteria and corresponding weight coefficients are set for different indicators such as vehicle status data (speed, collision time, etc.) and environmental data (traffic flow, road type, weather conditions, etc.) to build a risk assessment model:

[0075]

[0076] Among them, R represents the comprehensive risk score, x i is the score of indicator i, w i is the weight of indicator i, and n represents the number of indicators. Based on the risk percentage range, the scenarios are divided into three risk levels, namely normal, critical and dangerous scenarios. Scenario risk percentage R ％ The calculation formula is as follows:

[0077]

[0078] where R min Indicates the minimum risk score, R max Indicates the highest risk score.

[0079] Table 1. Risk classification

[0080]

[0081]

[0082] In step 1, the element information of each traffic scenario was extracted and standardized, and organized into a structured format. In step 2, based on the constructed risk assessment model, the key risk factors of each traffic scenario were quantitatively analyzed to obtain their risk levels, which were then incorporated into the structured description of each scenario.

[0083] Step 3. Based on the hierarchical classification results of the scenario elements and the preliminary risk level classification results in Step 2, association rule mining technology is used to mine the association rules between the scenario elements and risk levels, and a rule-based classifier is constructed to classify the risk level of the scenario to be assessed;

[0084] The association rule mining technology is Apriori algorithm;

[0085] Through the layer-by-layer iteration mechanism of the Apriori algorithm, the association between a single factor and multiple factor combinations on risk levels is discovered, forming an original rule base of factor combination-risk level;

[0086] Use the Apriori algorithm to calculate support, confidence, and lift, and filter rules that meet the threshold;

[0087] The support The calculation formula is as follows:

[0088]

[0089] Where X is the traffic factor, Y is the risk level, P(X∩Y) represents the probability of both X and Y, N(X∩Y) is the number of scenarios containing both X and Y, and N represents the total number of scenarios;

[0090] Support is used to measure the proportion of a certain type of traffic element and a specific risk level in all scenarios. By setting a support threshold, rules with higher predictive power can be screened out;

[0091] Confidence It represents the probability of scene Y appearing in a scene containing certain specific elements X. The calculation formula is as follows:

[0092]

[0093] Where P(Y|X) represents the probability of Y occurring under the condition that X occurs, and Support(X) represents the proportion of scenarios in which traffic element X occurs to the total number of scenarios. Confidence is used to assess the probability of a certain type of risk occurring when certain traffic elements occur. By setting a minimum confidence threshold, high-frequency and statistically significant scenarios are retained to avoid noise interference.

[0094] Lift It is used to measure the strength of the correlation between factors X and Y (more valuable than random guessing). The calculation formula is as follows:

[0095]

[0096] Where P(X) represents the probability of X appearing independently, P(Y) represents the probability of Y appearing independently, Support(X) represents the proportion of scenarios in which traffic factor X appears in the total number of scenarios, and Support(Y) represents the proportion of scenarios in which risk level Y appears in the total number of scenarios. Lift is used to reveal the dependency between factor combinations and risk levels, that is, whether the occurrence of a specific factor combination significantly affects the probability of a certain risk level, so as to identify rules that are truly causal or strongly correlated (Lift>1 indicates positive correlation, Lift=1 indicates independence, and Lift<1 indicates negative correlation).

[0097] Through the layer-by-layer iteration mechanism of the Apriori algorithm, it is found that the association between a single factor and multiple factor combinations on the risk level is formed (factor combination X * The original rule base for risk level Y);

[0098] The generated original rule base is cleaned and deduplicated, and weights are assigned to all cleaned rules. A rule-based classifier is constructed. Each scene feature element to be classified is matched with the rules in the classifier. The weight of each successfully matched rule is calculated, and a comprehensive score is calculated for each risk level. The calculation formula is:

[0099]

[0100] Where n is the number of matching rules, X * i is the i-th characteristic factor combination, Y is the risk level, and an aggressive strategy is used to divide the risk levels.

[0101] Adopting an aggressive strategy to categorize risk levels means:

[0102] The risk level with the highest comprehensive score is taken as the final classification result;

[0103] If the calculated high risk comprehensive score = low risk comprehensive score, the risk is classified as high.

[0104] The cleaning and deduplication process is as follows:

[0105] Merge duplicate rules: If two rules have the same content, keep the one with higher confidence.

[0106] Eliminate redundant rules: If the precondition of rule A is a subset of the precondition of rule B, and the two have the same conclusion, delete the rule with smaller coverage.

[0107] Conflict resolution: If two rules have the same preconditions but different conclusions, the rule with higher confidence is retained.

[0108] Step 4. Structured storage of scenario elements and risk levels to build an autonomous driving scenario library that supports multi-dimensional queries. Based on this library, technicians can flexibly select test sets with specific elements and risk levels based on different testing requirements to create corresponding autonomous driving test scenarios.

[0109] Build a data model to organize and store the multi-layer elements and risk levels of each scenario for complex queries and multi-dimensional analysis. The organizational storage method should support complex queries based on a single element or a combination of multiple elements, quickly locate specific test scenarios, and support analysis from different dimensions (such as risk level, scenario element type, etc.) to meet the complex needs of autonomous driving system testing. It can flexibly filter out test sets with specific elements and risk levels based on different testing needs to improve the targetedness and efficiency of testing. For example, testers can filter out scenes with specific elements (such as highways, night driving, and foggy weather) and medium risk levels from a variety of traffic scenarios for testing.

[0110] A computer-readable storage medium stores computer program instructions, which can be executed by a processor to implement the above-mentioned method for constructing an autonomous driving test scenario library based on real traffic data.

[0111] Here is a reference storage method:

[0112] Scenario: scenario number, risk level label, and scenario storage path.

[0113] Host vehicle: speed, yaw angle, vehicle type, etc.

[0114] Weather layer: visibility, rainfall, wind speed, etc.

[0115] Traffic participant layer: type, quantity, etc.

[0116] Temporary construction layer: construction signs, construction type, etc.

[0117] Transportation facilities layer: facility status, type, quantity, etc.

[0118] Road topology layer: road width, curvature, type, slope, etc.

[0119] The implementation principle of the method for constructing an autonomous driving test scenario library based on real traffic data of the present invention is as follows: by processing the multimodal sensor data collected in the traffic scene, the feature recognition and element extraction of the traffic scene are realized; combining the hierarchical structure analysis method of the scene elements, a risk assessment model is constructed to realize the preliminary risk level division; using the association rule mining algorithm to obtain the association rules between the scene elements and the risk level, and constructing a rule-based scene classifier to realize the automation of the scene risk classification; constructing an autonomous driving test scenario library based on the scene elements + risk level. The present invention can more comprehensively capture the characteristics of the traffic scene, more accurately identify the scene elements and scientifically divide the risk level of the scene, and combine the two to construct the scenario library, thereby improving the diversity and pertinence of the test scene, and providing more powerful support for the development and verification of autonomous driving technology.

[0120] This construction method not only improves the coverage of test scenarios, but also enables rapid retrieval of test scenarios with specific characteristics and risk levels based on different testing requirements. This will help improve the safety and stability of autonomous driving systems and accelerate their commercialization.

Claims

1. A method for constructing an autonomous driving test scenario library based on real traffic data, characterized in that: The steps include: Step 1. Based on the selective sensor fusion framework, dynamically adjust the fusion strategy of the multimodal sensor according to the current driving environment. The selective sensor fusion framework is configured as follows: There are multiple convolutional neural networks (CNNs) set up, and each sensor data input is sent to its corresponding convolutional neural network (CNN) to extract the data features F i , according to the data feature F i , identifying the context information of the current scene, and selecting a sensor fusion strategy suitable for the current scene based on the context information; using a preset deep learning model branch to process the corresponding sensor fusion strategy to obtain a target scene result and corresponding scene elements; Step 2. Normalize the information related to the scenario elements obtained in Step 1, then classify the scenario elements into layers, select key elements in each layer of the scenario as risk assessment indicators, build a risk assessment model, calculate the comprehensive risk score, and preliminarily divide the risk level; Step 3. Based on the hierarchical classification results of the scenario elements and the preliminary risk level classification results in Step 2, association rule mining technology is used to mine the association rules between the scenario elements and risk levels, and a rule-based classifier is constructed to classify the risk level of the scenario to be assessed; Step 4. Structured storage of scenario elements and risk levels to build an autonomous driving scenario library that supports multi-dimensional queries.

2. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: There are multiple deep learning model branches, each of which processes a specific sensor fusion strategy; The processing result of the deep learning model branch on the sensor fusion strategy includes at least a bounding box, a confidence level, and a type label of the target.

3. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: It also includes performing non-maximum suppression or weighted box fusion on the processing result of the deep learning model branch on the sensor fusion strategy.

4. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: Step 2: The scene elements are divided into six layers of scene elements: The main vehicle layer includes the status of the vehicle under test; The weather layer includes external environmental conditions that affect perception and vehicle dynamics; Traffic participant layer, including dynamic obstacle types and behavior intentions; Temporary construction layer, including non-standardized road change elements; Traffic facilities layer, including fixed traffic management and guidance facilities; The road topology layer includes the structural features of the road network.

5. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: The calculation formula for constructing the risk assessment model described in step 2 is: Among them, R represents the comprehensive risk score, x i is the score of risk assessment index i, w i is the weight of risk assessment indicator i, and n represents the number of risk assessment indicators; Scenarios are divided into three risk levels based on the scenario risk percentage range, namely normal, critical and hazardous scenarios. The risk levels are divided by percentage: normal (0% ≤ R < 30%), critical (30% ≤ R < 70%), and hazardous (R ≥ 70%). Scenario risk percentage R ％ The calculation formula is as follows: where R min Indicates the minimum risk score, R max Indicates the highest risk score.

6. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: In step 3, the association rule mining technology is the Apriori algorithm; Through the layer-by-layer iteration mechanism of the Apriori algorithm, the association between a single factor and multiple factor combinations on risk levels is discovered, forming an original rule base of factor combination-risk level; The generated original rule base is cleaned and deduplicated, and weights are assigned to all cleaned rules to obtain a rule-based classifier; For each scenario to be evaluated, the scenario elements are matched with the rules in the classifier. The weight of each successfully matched rule is calculated, and a comprehensive score is calculated for the risk level of the scenario. The calculation formula is: Where n is the number of matching rules, x i is the score of risk assessment indicator i, Y is the risk level, and an aggressive strategy is used to divide the risk levels.

7. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: The use of an aggressive strategy for risk classification refers to: The risk level with the highest comprehensive score is taken as the final classification result; If the calculated high risk comprehensive score = low risk comprehensive score, the risk is classified as high.

8. The method for constructing an autonomous driving test scenario library based on real traffic data according to claim 1, characterized in that: The multimodal sensor includes at least a camera and a laser radar.

9. A computer-readable storage medium, characterized in that Computer program instructions are stored, and the computer-readable instructions can be executed by a processor to implement the method for constructing an autonomous driving test scenario library based on real traffic data as described in any one of claims 1 to 8.

Images