A method and system for extracting a test case of a site based on feature points of natural driving scene data

Abstract

The site test case extraction method and system for natural driving scene data feature point selection belong to the technical field of vehicle road data collection, and solve the problem of key feature loss caused by Markov Monte Carlo and Gibbs test case extraction. The method comprises the following steps: step S1, extracting the collected scene data; step S2, extracting the candidate vehicle speed and vehicle speed interval of different road types, and determining the scene parameters according to different scenes; step S3, extracting the feature value and feature interval of each scene sequence parameter under the vehicle speed interval. The present application is suitable for extracting site test cases based on natural driving scene data feature points in the technical field of vehicle road data collection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle road data collection, in particular to a site test case extraction method and system based on natural driving scene data feature point selection. BACKGROUND

[0002] With the gradual development of domestic intelligent vehicles to L3 or L4 level, mainly involving congestion following, high-speed driving and valet parking ADAS functions. It is necessary to analyze the data of highway scenes and urban expressway congestion scenes, extract representative scenes for reproduction simulation in specific sites, and realize rapid, safe and effective development iteration testing through specific site real vehicle testing. However, the exponential explosion of vehicle data, including millions of scenes, makes it difficult to extract representative and effective scenes for special test testing.

[0003] Currently, Markov Monte Carlo and Gibbs sampling methods are used to extract data based on probability statistics, and different description parameters of different traffic participants in each scene are obtained, such as relative speed, absolute acceleration and relative distance. This method is effective when the sample size is large, and the test cases extracted can reflect the overall scene, but this method has defects when it comes to specific site real vehicle test cases, which can cause key feature loss and data set problems.

[0004] In summary, the existing Markov Monte Carlo and Gibbs extraction test case methods can cause key feature loss and data set problems.

[0005] Therefore, in the prior art, the extraction test case method can cause key feature loss and data set problems, such as patent document CN113076697A, which discloses a "typical driving condition construction method, related device and computer storage medium", which extracts and constructs a typical driving condition of a certain vehicle use at the smallest granularity by using Markov Monte Carlo simulation method, so as to achieve the purpose of accurately constructing a typical driving condition. The technical solution described in this patent document can only achieve the purpose of accurately constructing a typical driving condition, and does not explicitly provide a solution to the problem of key feature loss and data set caused by Markov-Monte Carlo extraction test cases. Patent document CN111007880A discloses "an extended target tracking method based on automobile radar", which uses Gibbs sampling for data association, and updates the multi-target PMBM posterior density when the association is successful, effectively improving the tracking accuracy. This patent document improves the tracking accuracy of automobile radar, and its analysis results cannot provide a solution to the problem of key feature loss and data set caused by Gibbs extraction test cases. SUMMARY

[0006] The present application solves the problem that the existing Markov Monte Carlo and Gibbs extraction test cases can cause key features to be lost.

[0007] The natural driving scene data feature point selection site test case extraction method provided by the present application comprises the following steps:

[0008] Step S1, extracting the collected scene data;

[0009] Step S2, extracting candidate vehicle speeds and vehicle speed intervals of different road types, and determining scene parameters according to different scenes;

[0010] Step S3, extracting and determining the scene sequence parameters of each feature value and the feature interval under the vehicle speed interval.

[0011] Further, in an embodiment of the present application, the scene parameters extracted in step S1 include:

[0012] The vehicle pose information, target vehicle pose information, road information, weather information, and illumination.

[0013] The vehicle pose information includes vehicle speed, vehicle acceleration, vehicle heading angle, and vehicle driving curve radius.

[0014] The target vehicle pose information includes target vehicle relative speed, target vehicle relative distance, target vehicle absolute speed, target vehicle relative acceleration, target vehicle absolute acceleration, and target vehicle heading angle.

[0015] The road information includes road type, road curvature, lane line type, lane line color, and median type.

[0016] The weather information includes rainfall, wind speed, and weather type.

[0017] The illumination includes illumination type, illumination intensity, and illumination direction.

[0018] Further, in an embodiment of the present application, the candidate vehicle speeds and vehicle speed intervals of different road types in step S2 are respectively:

[0019] The candidate vehicle speed is the highest frequency vehicle speed feature value under each road type, which is the candidate vehicle speed corresponding to the road type.

[0020] The vehicle speed interval is an interval formed by expanding 5 km / h above and below the upper and lower boundaries of the candidate vehicle speed.

[0021] Further, in one embodiment of the present application, the step S3 extracts each characteristic value and characteristic interval of the determined scene order parameter, wherein:

[0022] The continuous parameter characteristic value is the upper 10% quantile value, the lower 90% quantile value and the maximum value of the frequency distribution of the parameter distribution; the value of the discrete parameter is the characteristic value of the parameter;

[0023] The characteristic interval is an interval formed by expanding 10% of the upper and lower limits of each parameter characteristic value; the discrete characteristic value is the characteristic interval.

[0024] Further, in one embodiment of the present application, the step S3 specifically comprises:

[0025] Step S301, under one vehicle speed interval, each characteristic value and characteristic interval of the specified second order parameter are calculated;

[0026] Step S302, each characteristic interval of the second order parameter is extracted, and the characteristic value and the characteristic interval of the second order parameter are added according to the actual situation; under the second order characteristic interval, the same method is used to extract each characteristic value and characteristic interval of other specified parameters until the extraction of the characteristic value and the characteristic interval of all specified parameters is completed.

[0027] Step S303, under different road type speed intervals, each characteristic value and characteristic interval of all specified parameters are extracted until the extraction of all speed intervals is completed.

[0028] Further, in one embodiment of the present application, the specified parameter is a scene parameter selected by a user.

[0029] Further, in one embodiment of the present application, in the step S302, according to the actual situation:

[0030] When the scene parameter distribution interval covers the parameter distribution interval in the ADAS function definition of a model passenger car, the parameter interval specified in the function definition is used to recalculate the parameter distribution, and then the characteristic value and the characteristic interval are calculated; when the scene parameter distribution interval does not cover the parameter distribution interval in the ADAS function definition of a model passenger car, the extracted results are supplemented according to the parameters and parameter values given in the function definition.

[0031] When extracting a scene parameter characteristic value and a characteristic interval, if the extracted parameter characteristic value or characteristic interval does not contain the corresponding parameter value in the failure scene, the parameter value of the parameter corresponding to the failure scene is added to the extracted characteristic value, and each parameter value in the failure scene is used to expand the characteristic value and the characteristic interval.

[0032] The natural driving scene data feature point selection site test case extraction system disclosed by the application, the system comprises the following modules:

[0033] Module S1, for extracting the collected scene data;

[0034] Module S2, for extracting candidate vehicle speeds and vehicle speed intervals of different road types, and determining scene parameters according to different scenes;

[0035] Module S3, for extracting and determining scene sequence parameters each feature value and feature interval under the vehicle speed interval.

[0036] The computer readable storage medium disclosed by the application has a computer program stored thereon, and the computer program is executed by a processor to implement the steps of the above-mentioned method.

[0037] The computer device disclosed by the application comprises a memory and a processor, and the memory stores a computer program, and when the processor runs the computer program stored in the memory, the steps of the above-mentioned method are executed.

[0038] The application solves the problem of key feature loss and data set caused by the existing Markov Monte Carlo and Gibbs extraction test case method. The specific beneficial effects include:

[0039] 1、The application realizes the extraction of site test cases with key features from millions of scenes by the objective data distribution feature points of traffic participants in different scenes, shortens the order of magnitude to hundreds, not only effectively solves the problem of key feature loss, but also makes the site test efficient and convenient.

[0040] 2、The application effectively solves the problem of the existing technology data set by extracting continuous parameters as the upper 10% quantile value, the lower 90% quantile value and the maximum frequency distribution.

[0041] 3、The application proposes a sampling method based on feature point extraction, which can dynamically increase feature points according to actual conditions to ensure the rationality of the extraction result.

[0042] The application is suitable for the field of intelligent automobile L3 / L4 level natural driving data key scene extraction technology, and proposes a natural driving scene data feature point selection site test case extraction method, which realizes the extraction of representative site test cases from millions of scenes by the objective data distribution feature points of traffic participants in different scenes, shortens the order of magnitude to hundreds, makes the site test efficient and convenient, and solves the technical problems of key feature loss, data set and large sample size of traditional sampling methods in the face of specific site real vehicle test cases. BRIEF DESCRIPTION OF DRAWINGS

[0043] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings of which:

[0044] Figure 1 is a scene classification graph according to the first embodiment.

[0045] Figure 2 is an extracted data graph according to the third embodiment.

[0046] Figure 3 is a test case extraction flow chart according to the fourth embodiment.

[0047] Figure 4 is a scene parameter extraction flow chart according to the eighth embodiment.

[0048] Figure 5 is a test case extraction result graph according to the eighth embodiment. DETAILED DESCRIPTION

[0049] Various embodiments of the present application will be described hereinafter with reference to the accompanying drawings. The embodiments described by reference to the drawings are exemplary and are intended to explain the present application, and should not be understood as limiting the present application.

[0050] In the first embodiment, the site test case extraction method for selecting feature points of natural driving scene data includes the following steps:

[0051] Step S1, extracting the collected scene data;

[0052] Step S2, extracting candidate vehicle speeds and vehicle speed intervals of different road types, and determining scene parameters according to different scenes;

[0053] Step S3, extracting and determining the feature values and feature intervals of the scene sequence parameters under the vehicle speed interval.

[0054] In the first embodiment, in step S1, traffic participant information is collected for different roads, different time periods, and different regions, and the scenes in the interaction process between the host vehicle and other traffic participants encountered during natural driving are classified. A typical classification method is shown in FIG. Figure 1

[0055] In the second embodiment, the site test case extraction method for selecting feature points of natural driving scene data according to the first embodiment is further limited. In the second embodiment, the condition for extracting the collected scene data in step S1 is:

[0056] ​According to the speed interval to which the relative speed between the ego vehicle and the target vehicle belongs, an interaction state in which the ego vehicle and the target vehicle are located is defined corresponding to the speed interval.

[0057] Specifically, the initial actions of the ego vehicle and the target vehicle are both following a line, when the relative speed is less than 0 km / h, the ego vehicle is in a state of chasing the front vehicle, when the relative speed is greater than 0 km / h, the ego vehicle is in a state of approaching the front vehicle, and when the interval between two consecutive same states is less than 1 s, the two states are merged.

[0058] Embodiment three, the embodiment is a further limitation of the natural driving scene data feature point selection site test case extraction method of embodiment one, in the embodiment, the scene parameters extracted from the collected scene data in step S1 include:

[0059] The ego vehicle pose information, the target vehicle pose information, the road information, the weather information and the illumination.

[0060] The ego vehicle pose information includes the ego vehicle speed, the ego vehicle acceleration, the ego vehicle heading angle and the ego vehicle driving curve radius;

[0061] The target vehicle pose information includes the target vehicle relative speed, the target vehicle relative distance, the target vehicle absolute speed, the target vehicle relative acceleration, the target vehicle absolute acceleration and the target vehicle heading angle;

[0062] The road information includes the road type, the road curvature, the lane line type, the lane line color and the median type;

[0063] The weather information includes the rainfall, the wind speed and the weather type;

[0064] The illumination includes the illumination type, the illumination intensity and the illumination direction.

[0065] Specifically, as shown in Figure 2 According to the different relative speeds, a computer program is written to extract the scene, and the collected data is extracted, and the extracted data should include the ego vehicle pose information, the target vehicle pose information, the road information, the weather and illumination information.

[0066] Embodiment four, the embodiment is a further limitation of the natural driving scene data feature point selection site test case extraction method of embodiment three, in the embodiment, the first order of the extracted scene parameters is the ego vehicle speed.

[0067] In the embodiment, the ego vehicle driving speed is determined as the first parameter, the user can select other scene parameters in addition to the ego vehicle speed, and the selected other scene parameters are sorted according to the importance, and the characteristic value and the characteristic interval of the extracted scene parameters are extracted. The extraction process of the test case is as follows Figure 3As shown, a parameter-based feature value extraction method is adopted. By analyzing the data distribution feature points in different scenarios, representative site test cases can be extracted from millions of scenarios, reducing the number of cases to a few hundred, making site testing efficient and convenient.

[0068] Specifically, the mathematical method for scene parameter extraction is as follows: Determine the initial value X. (1) That is, assuming that the sample x has already been obtained based on the distribution of the parameters to be extracted. (i) Let the next sample be x. (+1) =(x1) (i+1) x2 (i+1) x3 (i+1) ,…,x n (i+1) Therefore, it can be viewed as a vector, and for a certain component x... j (i+1) Extraction can be performed by obtaining the probability distribution of a component given that other components are known, and then combining this with characteristic values ​​such as parameters, probability distribution extrema, and quantiles in product failure scenarios. This is achieved through sample X. (+1) The component x1 already obtained (i+1) To x j-1 (i+1) To obtain the conditional probability distribution of the next component, i.e., P(x j-1 (i+1) |x1 (i+1) x2 (i+1) x3 (i+1) ,…,x n (i+1) Repeat the above process until all components have been extracted, and you will get a test case.

[0069] Implementation Method 5: This implementation method further defines the site test case extraction method for selecting feature points in natural driving scenario data as described in Implementation Method 1. In this implementation method, the candidate vehicle speeds and speed ranges for different road types mentioned in step S2 are as follows:

[0070] The candidate vehicle speed is the vehicle speed feature value with the highest frequency under each road type.

[0071] The speed range is the range formed by expanding the upper and lower limits of the candidate speed by 5 km / h.

[0072] In this embodiment, the most frequently occurring vehicle speed is used as the candidate speed because high-frequency speeds are representative. The upper and lower limits of the candidate speed are expanded by 5 km / h to form a speed range because the vehicle speed during operation is within ±5 km / h.

[0073] Embodiment six, the embodiment is further limited to the natural driving scene data feature point selection of the site test case extraction method described in embodiment one, in the embodiment, each feature value of the scene order parameter determined in step S3 is respectively:

[0074] The continuous parameter feature value is the upper 10% quantile value, the lower 90% quantile value, and the maximum value of the frequency distribution of the parameter distribution; the value of the discrete parameter is the feature value of the parameter;

[0075] The feature interval is an interval formed by expanding the upper and lower boundaries by 10% with each parameter feature value as the center; the discrete feature value is the feature interval.

[0076] Specifically, the continuous variable is converted into a non-continuous variable according to the requirements, for example, the illumination is defined as strong and weak in different distribution intervals of lx values. The continuous variable will effectively solve the problems in the existing technical data set through this extraction method.

[0077] Embodiment seven, the embodiment is further limited to the natural driving scene data feature point selection of the site test case extraction method described in embodiment one, in the embodiment, the parameter feature value is selected as:

[0078] If the corresponding next order parameter data is less than 4, the next order parameter value is selected as the feature value under the feature interval;

[0079] If the corresponding next order parameter data is greater than or equal to 4, the continuous parameter feature value is the upper 10% quantile value, the lower 90% quantile value, and the maximum value of the frequency distribution of the parameter distribution; the value of the discrete parameter is the feature value of the parameter.

[0080] In the embodiment, the upper 10% quantile value and the lower 90% quantile value are extracted as the feature value, the purpose is to extract the boundary of the parameter distribution and ensure its effectiveness, avoid the influence of the maximum and minimum value extraction method, and select the maximum value of the frequency distribution to ensure that the most representative parameter value is extracted and the representativeness of the final test case is ensured.

[0081] Embodiment eight, the embodiment is further limited to the natural driving scene data feature point selection of the site test case extraction method described in embodiment one, in the embodiment, step S3 specifically includes:

[0082] Step S301, under a vehicle speed interval, the feature value and feature interval of the specified second order parameter are calculated;

[0083] Step S302, each feature interval of the second order parameter is extracted, and the feature value of the second order parameter is added according to the actual situation. In the second order feature interval, the same method is used to extract each feature value and feature interval of other specified parameters until the feature value and feature interval of all specified parameters are extracted.

[0084] Step S303, under different road type speed intervals, extract each feature value and feature interval of all specified parameters until all speed intervals are extracted.

[0085] Specifically, the scene parameter extraction flowchart is as shown in Figure 4 which clearly and clearly shows the specific process of scene parameter extraction.

[0086] As shown in Figure 5 , after the end of the fixed parameter extraction process, a test case tree is constructed. A binary tree type test case tree is formed. Among them, ● represents the natural driving scene feature point; ▲ represents the product function definition feature point; ■ represents the natural driving scene feature interval; ◆ represents the product function definition feature interval; represents the failure case feature point; represents the failure case feature interval; ○ represents the extraction termination node; → represents the test case. In the figure, ● represents the feature point of continuous or discrete parameter distribution calculation, and represent additional feature points added by humans according to actual situations.

[0087] Embodiment nine, this embodiment is a further limitation of the natural driving scene data feature point selection site test case extraction method of embodiment eight. In this embodiment, the specified parameter is a scene parameter selected by the user.

[0088] In this embodiment, the scene parameters selected by the user include vehicle pose information, target vehicle pose information, road information, weather information, and illumination. Among them, the vehicle pose information includes vehicle speed, vehicle acceleration, vehicle heading angle, and vehicle driving curve radius; the target vehicle pose information includes target vehicle relative speed, target vehicle relative distance, target vehicle absolute speed, target vehicle relative acceleration, target vehicle absolute acceleration, target vehicle heading angle; the road information includes road type, road curvature, lane line type, lane line color, and isolation belt type; the weather information includes rainfall, wind speed, and weather type; the illumination includes illumination type, illumination intensity, and illumination direction. The user can select the scene parameters according to the requirements, and sort them according to the importance of the scene parameters to calculate the feature value and feature interval of each parameter.

[0089] Embodiment ten, this embodiment is a further limitation of the natural driving scene data feature point selection site test case extraction method of embodiment eight, in this embodiment, the step S302 is according to the actual situation for:

[0090] When the scene parameter distribution interval covers the parameter distribution interval in the definition of the ADAS function of a model passenger car, the parameter distribution is recalculated using the parameter interval specified in the function definition, and the characteristic value and the characteristic interval are calculated; when the scene parameter distribution interval does not cover the parameter distribution interval in the definition of the ADAS function of a model passenger car, the extracted results are supplemented according to the parameters and parameter values given in the function definition.

[0091] When extracting a scene parameter characteristic value and a characteristic interval, if the extracted parameter characteristic value or characteristic interval does not contain the corresponding parameter value in the failure scene, the parameter value corresponding to the failure scene should be added to the extracted characteristic value, and each parameter value in the failure scene should be used to expand the characteristic value and the characteristic interval.

[0092] Embodiment eleven, the natural driving scene data feature point selection site test case extraction system of this embodiment, the system includes the following modules:

[0093] Module S1, for extracting the collected scene data;

[0094] Module S2, for extracting candidate vehicle speeds and vehicle speed intervals of different road types, and determining scene parameters according to different scenes;

[0095] Module S3, for extracting and determining each characteristic value and characteristic interval of the scene sequence parameter under the vehicle speed interval.

[0096] Embodiment twelve, a computer readable storage medium, a computer program is stored on the computer readable storage medium, when the computer program is executed by the processor, the steps of the method described in any one of embodiments one to ten are implemented.

[0097] Embodiment thirteen, a computer device, comprising a memory and a processor, the memory stores a computer program, when the processor runs the computer program stored in the memory, the steps of the method described in any one of embodiments one to ten are executed.

Claims

1. A method for extracting site test cases from feature points of natural driving scenario data, characterized in that, It comprises the following steps: Step S1, extracting the collected scene data; Step S2, extracting the candidate vehicle speed and vehicle speed interval of different road types, and determining the scene parameters according to different scenes; Step S3, extracting the characteristic value and characteristic interval of each feature value of the determined scene sequence parameters under the vehicle speed interval; The step S3 specifically comprises: Step S301, calculating the characteristic value and characteristic interval of each feature value of the specified second sequence parameter under a vehicle speed interval of a host vehicle; Step S302, extracting the characteristic value and characteristic interval of each feature value of the second sequence parameter, adding the characteristic value and characteristic interval of the second sequence parameter according to the actual situation, extracting the characteristic value and characteristic interval of each feature value of other specified parameters under the second sequence characteristic interval by using the same method, and until the extraction of the characteristic value and characteristic interval of all specified parameters is completed; Step S303, extracting the characteristic value and characteristic interval of each feature value of all specified parameters under the vehicle speed interval of different road types, and until the extraction of all vehicle speed intervals is completed; The step S302 according to the actual situation comprises: When the scene parameter distribution interval covers the parameter distribution interval in the ADAS function definition of a model passenger car, the parameter interval in the function definition is used to recalculate the parameter distribution, and then the characteristic value and characteristic interval are calculated; when the scene parameter distribution interval does not cover the parameter distribution interval in the ADAS function definition of a model passenger car, the extracted results are supplemented according to the parameters and parameter values given in the function definition; When extracting the characteristic value and characteristic interval of a scene parameter, if the extracted parameter characteristic value or characteristic interval does not contain the corresponding parameter value in the invalid scene, the parameter value corresponding to the parameter in the invalid scene should be added to the extracted characteristic value, and each parameter value in the invalid scene is used to expand the characteristic value and characteristic interval.

2. The natural driving scene data feature point selection site test case extraction method according to claim 1, characterized in that, The scene parameters extracted by the step S1 comprise: The host vehicle pose information, the target vehicle pose information, the road information, the weather information and the illumination; The host vehicle pose information comprises the host vehicle speed, the host vehicle acceleration, the host vehicle heading angle and the host vehicle driving curve radius; The target vehicle pose information comprises the target vehicle relative speed, the target vehicle relative distance, the target vehicle absolute speed, the target vehicle relative acceleration, the target vehicle absolute acceleration and the target vehicle heading angle; The road information comprises the road type, the road curvature, the lane line type, the lane line color and the separation belt type; The weather information comprises the rainfall, the wind speed and the weather type; The illumination comprises the illumination type, the illumination intensity and the illumination direction.

3. The natural driving scene data feature point selection site test case extraction method according to claim 1, characterized in that, The candidate vehicle speed and vehicle speed interval of different road types in the step S2 are as follows: The candidate vehicle speed is the highest frequency vehicle speed characteristic value under each road type, which is used as the candidate vehicle speed corresponding to the road type; The vehicle speed interval is an interval formed by expanding 5km / h above and below the center of the candidate vehicle speed.

4. The natural driving scene data feature point selection site test case extraction method according to claim 1, characterized in that, The characteristic value and characteristic interval of each feature value of the determined scene sequence parameters in the step S3 are as follows: The continuous parameter characteristic value is the upper 10% quantile value, the lower 90% quantile value and the maximum value of the frequency distribution of the parameter distribution; the value of the discrete parameter is the characteristic value of the parameter; The characteristic interval is an interval formed by expanding 10% of the upper and lower limits of each parameter characteristic value as the center.

5. The natural driving scene data feature point selection site test case extraction method according to claim 1, characterized in that, The specified parameter is a scene parameter selected by a user.

6. A test case extraction system for natural driving scene data feature point selection, characterized by, The system is based on the natural driving scene data feature point selection site test case extraction method of claim 1, and the system comprises the following modules: Module S1 is used for extracting the collected scene data; Module S2 is used for extracting the candidate vehicle speed and vehicle speed interval of different road types, and determining the scene parameters according to different scenes; Module S3 is used for extracting and determining the characteristic value of each feature value and the characteristic interval of the scene sequence parameter under the vehicle speed interval.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 5. 8.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-7. When the processor runs the computer program stored in the memory, the steps of the method of any one of claims 1 to 5 are executed.

Images