Intelligent driving car open road test system and use method

Abstract

The application relates to an intelligent driving automobile open road test system and a use method, and belongs to the technical field of open road test of intelligent driving automobiles. The problems that test time is long when an intelligent driving automobile is tested on an open road, and test for some scenes is not sufficient are solved. The open road test system optimizes test periods and road scene coverage degrees cooperatively, fully tests possible scenes, reduces the repetition times of the same test scenes, and reduces test periods.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of open road testing of intelligent driving cars, and particularly relates to an open road testing system for intelligent driving cars and a use method thereof. BACKGROUND

[0002] An open road refers to a road on which social vehicles and pedestrians pass, and the scene is randomly variable and the traffic condition is complex. Open road testing is the final link of road-in-loop testing and is also a necessary link for an automatic driving vehicle to complete testing. Open road testing can more comprehensively and realistically test the running condition of an automatic driving vehicle under various complex scene conditions and the resolution of crisis situations, and comprehensively test the running, system working condition, module function, and body feeling of the vehicle in each dimension.

[0003] In the prior art, there is no theoretical system method for guiding open road testing. Most theoretical testing schemes usually set the vehicle in a closed testing environment and sequentially test different driving scenes. The traffic scenes occurring in the testing process are mainly set by people, and since the process of setting scenes by people depends on the driving experience of the tester and the collection of classic scenes, the closed road testing is difficult to cover a variety of complex conditions that can occur in the actual driving environment. And compared with the closed driving environment, the time node of the scene occurring in the actual driving environment has uncertainty, which further increases the challenge to the scene recognition ability of the intelligent driving car or intelligent driving system. Therefore, the closed road testing is not enough to prove the safe driving ability of the intelligent driving car. At the same time, due to the rich traffic scenes, complex traffic conditions, and uncertain time of scene occurrence in the actual driving environment, the open road testing has the disadvantages of low testing efficiency, long testing period, and poor scene coverage.

[0004] For example, Chinese patent CN115541248A discloses a test system and test method of a driving assistance system, mainly providing a test system using multiple sensors to avoid the influence of social vehicles. Through the fusion perception of multiple sensors, the perception ability of the vehicle to the surrounding environment is improved, so that more accurate target vehicles are obtained during the driving test. However, this patent focuses on improving the perception ability of the vehicle and cannot solve the problems of long test period and low scene coverage of open road testing. Chinese patent CN114753199A discloses an open road grading method and device based on intelligent networked vehicle testing. The invention grades the intelligent networked degree of open roads according to the complexity of the current road, traffic complexity, environmental impact degree, and the degree of basic digital infrastructure construction. The invention realizes the grading processing of open roads and provides a theoretical basis for selecting roads for open road testing of intelligent vehicles. Since the invention focuses more on road grading, open road testing involves factors beyond just road types, so it does not comprehensively solve the problem of improving scene coverage.

[0005] Therefore, in order to solve the problem of long test period and insufficient testing of certain scenes for intelligent driving vehicles during open road testing, an intelligent driving vehicle open road testing system and use method are needed. SUMMARY

[0006] In view of the above problems, the present application provides an intelligent driving vehicle open road testing system and use method, in particular a test period and road scene coverage cooperative optimization technology, which solves the problem of complex traffic scenes and long test period in the prior art open road. The test method of the present application can fully test possible scenes and reduce the number of repeated tests of the same test scene, thereby reducing the test period.

[0007] In order to achieve the above purpose, the present application provides the following scheme:

[0008] A use method of an intelligent driving vehicle open road testing system, comprising:

[0009] S1, collecting scene information of an open road section based on traffic big data and road network information; classifying the scene information of the open road section;

[0010] S2, inputting the classified scene information of step S1 into a simulation scene for screening, constructing a temporary scene library set A from the screened scenes, and building a test site based on the temporary scene library set A and scene coverage;

[0011] S3, selecting an experimental vehicle to test in a test section of the test site described in step S2, identifying the scene of the temporary scene library described in step S2 in the test section, collecting the scene data information, performing scene understanding and classification on the scene data information;

[0012] S4, obtaining the single test time of the experimental vehicle in the open road test section in step S3 and the test route corresponding to the single test time, arranging and optimizing the test route and test time according to the scene data information in step S3, and establishing the optimal test section of the intelligent driving vehicle on the open road.

[0013] Preferably, the specific steps of classifying the scene information of the open road section in step S1 include:

[0014] S11, selecting a test site for open road testing of the experimental vehicle, and the test site is an actual traffic road;

[0015] S12, dividing the test section described in step S11 into special driving sections and non-special driving sections according to the data provided by the road network information;

[0016] Among them, the special driving section includes but is not limited to: the road around the hospital, the road around the school, and the road around the business circle;

[0017] S13, collecting the scene information of the test open road section described in step S11 based on the traffic big data and the road network information; using the test road scene information included in the special driving section described in step S12 to filter the open road scene information, and dividing the scene information of the open road section into special driving section scene information and non-special driving section scene information.

[0018] Further, the road network information described in step S1 includes but is not limited to: building information, traffic stations, natural landscapes, traffic accidents, waterway and railway information.

[0019] Further, the road network information data provides the road network topology, road type, intersection geometry and road surface marking of the road; the topology includes but is not limited to: bus type road structure, grid type road structure, tree type road structure and composite type road structure; the road type includes but is not limited to: main road, main and subordinate lane, overtaking lane and ramp; the intersection geometry includes but is not limited to: cross intersection, T-shaped intersection and roundabout intersection; the road surface marking includes but is not limited to: pedestrian crossing, no parking line and indicating arrow.

[0020] Preferably, the traffic big data described in step S1 includes:

[0021] a. Static data, including but not limited to: road level, road attribute, road width, road speed limit, road lane number and whether the road is closed;

[0022] b. Dynamic data, including but not limited to: road traffic speed, event information, traffic volume, weather, visibility, pedestrian situation, non-motor vehicle situation, signal light and traffic density.

[0023] The technical scheme of the present application is based on the acquisition and analysis of traffic big data information, further obtains multi-dimensional related data affecting traffic scene screening and traffic scene identification, which completely covers the factors that need to be considered in the traffic scene screening process, and provides help for detailed and accurate scene screening.

[0024] Preferably, the specific steps of building the test site in step S2 include:

[0025] S21, input the special driving section scene information and the non-special driving section scene information in step S13 into the simulation scene for screening to obtain the screened scene;

[0026] S22, establish a temporary scene library set A according to the screened scene in step S21, and build a test site based on the temporary scene library set A and the temporary scene coverage, wherein A={A i |A1,A2,…A n}, i=1,2,3…,n, set the scene set of the complete scene library as Ω, and obtain the temporary scene coverage of the open road test section based on the scene set A of the temporary scene library and the scene set Ω of the complete scene library.

[0027] Preferably, the simulation scene library in step S21 is an OpenScenario scene library.

[0028] Further, the screening method in step S21 is to match and screen the scene occurring in the corresponding road information through the reference directory in the simulation scene library, and the reference directory is a road network reference.

[0029] Preferably, the number of scene repetitions of the scene set A of the temporary scene library in step S22 is:

[0030]

[0031] Wherein, R is the number of scene repetitions of the scene set A of the temporary scene library; A i is the i-th temporary scene library in the scene set A, and A jis the jth temporary scene library in the scene set A, i≠j; when c=1, that is, the case of covering all scenes in all scene libraries, the temporary scene library with the minimum number of repeated scenes in the scene set A of the temporary scene library is selected as the final test road section of the test site, and the corresponding road section is selected as the final test road section of the test site;

[0032] Preferably, the expression of the temporary scene coverage in step S22 is:

[0033]

[0034] Wherein, c is the temporary scene coverage, 0≤c≤1; Ω is the scene set of the complete scene library, A is the scene set of the temporary scene library, A=|A1∪A2...∪A n |.

[0035] In the technical scheme of the present application, when all scenes in all scene libraries are covered, the environment with the minimum number of repeated scenes can be selected for open road testing through the above formula. In addition, the present application narrows down the less and more accurate traffic scenes to the range of scene recognition in the next step, avoids the interference of irrelevant traffic scenes on the identification of the target traffic scene, and further improves the accuracy of traffic scene identification.

[0036] Preferably, the specific steps of the test vehicle in the test road section of the test site in step S2 are as follows:

[0037] S31, preparing a test vehicle, the test vehicle is an intelligent driving vehicle, and a laser radar and a camera are installed on the test vehicle;

[0038] S32, testing a person to drive the test vehicle in step S31 to pass through the open road section to be tested in step S1, repeatedly driving on the open road section, and collecting scene data information of the open road section, the scene of the open road section being the scene in the temporary scene library in step S2;

[0039] S33, after the scene data information collection in step S32 is completed, the scene data information is understood and classified, and the classified scene data includes four dimensions: scene category, scene occurrence time, scene end time and scene occurrence location.

[0040] Preferably, the laser radar in step S31 is arranged on the roof of the test vehicle to collect point cloud data during driving; the camera is a binocular camera arranged at the front and rear of the test vehicle, pointing to the driving direction of the test vehicle and collecting image data.

[0041] Preferably, the data information in step S32 includes image information collected by the camera and point cloud information collected by the laser radar.

[0042] Preferably, the repeated experiments in step S32 collect data at a fixed frequency.

[0043] Preferably, the classification in step S33 is a static scene classification technique, including: based on deep learning or clustering algorithm.

[0044] The technical scheme of the present application repeats experiments on the same open road multiple times, better adapts to the random traffic scenes occurring in the open road, and improves the scene coverage of data collection.

[0045] Preferably, the specific steps of the arrangement optimization in step S4 include:

[0046] S41, obtaining the single test time of the experimental vehicle on the open road test section and the test route corresponding to the single test time in step S32;

[0047] S42, respectively inputting the scene classification data obtained in step S33 and the test route corresponding to the single test time in step S41 into a machine learning algorithm, using the machine learning algorithm to analyze the time and space distribution relationship model of the specific scene type of the test route corresponding to the single test time, and obtaining the scene quantity of the test route corresponding to the single test time;

[0048] S43, obtaining the optimal test section of the intelligent driving car on the open road based on the scene quantity of the test route corresponding to the single test time in step S42 and the loss function formula.

[0049] Preferably, the loss function formula of the special road in step S43 is:

[0050]

[0051] Wherein: x n＊a represents the current point at each iteration on the n-th special road longitude, x n＊b represents the next special road point selected at each iteration on the n-th special road longitude, y n＊a represents the current point at each iteration on the n-th special road dimension, y n＊b represents the next special road point selected at each iteration on the n-th special road dimension, num represents the scene quantity of the test route corresponding to the single test time.

[0052] In the present application, the optimal test route is set as the test route with the lowest loss function, which ensures that the test time period is reduced without losing the scene coverage as much as possible. By obtaining the test time and test route of the test vehicle, efficient, high-coverage, and highly applicable open road test of intelligent vehicles is completed.

[0053] Compared with the prior art, the present application has at least the following beneficial effects:

[0054] The present application fully combines road classification, perception algorithm, situation awareness and scene classification method, cooperative optimization method, guides the whole test process, fully considers various factors leading to long scene coverage and test period in the actual intelligent driving vehicle test process, provides theoretical guidance for open road test of all intelligent driving vehicles, helps to shorten the test period, and the reduction of the test period is more significant considering the increase of experimental vehicles, has high theoretical value and practical significance. BRIEF DESCRIPTION OF DRAWINGS

[0055] The accompanying drawings are only for the purpose of illustrating specific embodiments and are not considered to be limiting of the present application.

[0056] Figure 1 A flowchart of the open road test system of the intelligent driving vehicle of the present application;

[0057] Figure 2 A principle diagram of the complete test period and scene coverage cooperative optimization technology of the present application. DETAILED DESCRIPTION

[0058] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict. In addition, the present application can also be implemented in other ways different from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0059] One specific embodiment of the present application, such as Figure 1 , discloses an intelligent driving vehicle open road test system and a use method, comprising:

[0060] S1, collecting scene information of an open road section based on traffic big data and road network information; classifying the scene information of the open road section;

[0061] S2, inputting the classified scene information of step S1 into a simulation scene for screening, constructing a temporary scene library set A from the screened scenes, and building a test site based on the temporary scene library set A and scene coverage;

[0062] S3, selecting an experimental vehicle to test in the test section of the test site of step S2, identifying the scenes of the temporary scene library of step S2 in the test section, collecting scene data information, and classifying the scene data information through scene understanding;

[0063] S4, obtaining the single test time of the experimental vehicle on the open road test section in step S3 and the test route corresponding to the single test time, arranging and optimizing the test route and test time according to the scene data information classification in step S3, and establishing the optimal test section of the intelligent driving vehicle on the open road.

[0064] Preferably, the specific steps of classifying the scene information of the open road section in step S1 include:

[0065] S11, selecting a test site for open road testing of the experimental vehicle, and the test site is an actual traffic road;

[0066] S12, classifying the test section in step S11 into a special driving section and a non-special driving section according to the data provided by the road network information;

[0067] The special driving section includes but is not limited to: roads around hospitals, schools and business circles;

[0068] S13, collecting the scene information of the test open road section in step S11 based on the traffic big data and the road network information; using the test road scene information included in the special driving section in step S12 to filter the open road scene information, and classifying the scene information of the open road section into special driving section scene information and non-special driving section scene information.

[0069] Further, the road network information in step S1 includes but is not limited to: building information, traffic sites, natural scenery, traffic accidents, waterway and railway information.

[0070] Further, the road network information data provides the road network topology, road type, intersection geometry and road surface marking of the road; the topology includes but is not limited to: bus type road structure, grid type road structure, tree type road structure and composite type road structure; the road type includes but is not limited to: trunk road, master-slave lane, overtaking lane and ramp; the intersection geometry includes but is not limited to: cross intersection, T-shaped intersection and roundabout intersection; the road surface marking includes but is not limited to: pedestrian crossing, no parking line and indication arrow.

[0071] Preferably, the traffic big data in step S1 includes:

[0072] a. Static data, including but not limited to: road grade, road attribute, road width, road speed limit, road lane number and whether the road is closed;

[0073] b. Dynamic data, including but not limited to: road traffic speed, event information, traffic volume, weather, visibility, pedestrian condition, non-motor vehicle condition, signal light and traffic density.

[0074] The technical solution of the present application is based on the acquisition and analysis of traffic big data information, and further obtains multi-dimensional related data affecting traffic scene screening and traffic scene identification. These data completely cover the factors that need to be considered in the traffic scene screening process, and provide help for detailed and accurate scene screening.

[0075] Preferably, the specific steps of building the test site in step S2 include:

[0076] S21, matching the special driving section scene information and the non-special driving section scene information in the screening step S12 in the road network reference directory of the OpenScenario scene library;

[0077] S22, establishing a temporary scene library set A according to the screened scene in step S21, and building a test site based on the temporary scene library set A and the temporary scene coverage, wherein A={A i |A1,A2,…A n}, i=1, 2, 3…, n, and setting the scene set of the complete scene library as Ω, obtaining the temporary scene coverage of the open road test section based on the scene set A of the temporary scene library and the scene set Ω of the complete scene library.

[0078] Preferably, the number of scene repetitions of the scene set A of the temporary scene library in step S22 is:

[0079]

[0080] Wherein, R is the number of scene repetitions of the scene set A of the temporary scene library; A i is the i-th temporary scene library in the scene set A, and A j is the j-th temporary scene library in the scene set A, i≠j; when c=1, that is, all scenes in the complete scene library are covered, the temporary scene library with the smallest number of scene repetitions R of the scene set A of the temporary scene library and the corresponding road section are selected as the final test section of the test site.

[0081] Preferably, the expression of the temporary scene coverage in step S22 is:

[0082]

[0083] Wherein, c is the temporary scene coverage, 0≤c≤1; Ω is the scene set of the complete scene library, A is the scene set of the temporary scene library, A=|A1∪A2...∪A n |.

[0084] Preferably, the specific steps of the test vehicle in step S3 testing the test section of the test site in step S2 are as follows:

[0085] S31, prepare an experimental vehicle, the experimental vehicle is an intelligent driving vehicle, install a laser radar and a camera on the experimental vehicle;

[0086] S32, test the experimental vehicle in step S31 driven by a person to pass through the open road section to be tested in step S1, repeatedly drive on the open road section, and collect scene data information of the open road section, the scene of the open road section is the scene in the temporary scene library in step S2.

[0087] S33, after the completion of the scene data information collection in step S32, the scene data information is understood and classified, the classified scene data includes four dimensions: scene category, scene occurrence time, scene end time and scene occurrence place.

[0088] Preferably, the laser radar in step S31 is arranged on the roof of the experimental vehicle, and point cloud data in the driving process is collected; the camera is a binocular camera, which is arranged at the front and rear of the experimental vehicle and points to the driving direction of the experimental vehicle and collects image data.

[0089] Preferably, the data information in step S32 includes image information collected by the camera and point cloud information collected by the laser radar.

[0090] Preferably, in step S32, the data collection is selected at a fixed frequency.

[0091] Preferably, the classification in step S33 is a static scene classification technology, which includes: based on deep learning or clustering algorithm.

[0092] The technical scheme of the present application is repeated in the same section of the open road, which better adapts to the random traffic scenes in the open road and improves the scene coverage of data collection.

[0093] Preferably, the specific steps of step S4 include:

[0094] S41, obtain the single test time of the experimental vehicle in step S32 on the open road test section and the test route corresponding to the single test time;

[0095] S42, respectively, the scene classification data obtained in step S33, the test route corresponding to the single test time in step S41 is input into the machine learning algorithm, the machine learning algorithm is used to analyze the time and space distribution relationship model of the specific scene type of the test route corresponding to the single test time, and the scene quantity of the test route corresponding to the single test time is obtained.

[0096] S43, based on the special driving section in step S12, the special road of the single test section in step S41 is numbered {1, 2, 3..., n}, the special road site number, the special road point coordinate and the special road point scene number are collected respectively; the loss function of the special road is constructed based on the special road point coordinate; the scene quantity in the test route corresponding to the single test time in step S42, the loss function of the special road, the special road site number and the special road point scene number are input into the simulated annealing algorithm for calculation and analysis, and a route which can pass through all the special road points is output.

[0097] Preferably, the loss function formula of the special road in step S43 is:

[0098]

[0099] Wherein: x n＊a represents the current point of each iteration on the n-th special road longitude, x n＊b represents the next special road point selected at each iteration on the n-th special road longitude, y n＊a represents the current point of each iteration on the n-th special road latitude, y n＊b represents the next special road point selected at each iteration on the n-th special road latitude, and num represents the scene quantity of the test route corresponding to the single test time.

[0100] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A method for using an open road testing system for intelligent driving vehicles, characterized in that... Comprise: S1, based on traffic big data and road network information collection open road section of the scene information; the open road section of the scene information is classified; S2, input the classified scene information in step S1 into a simulation scene for screening, and construct a temporary scene library collection from the screened scenes A , build a test site based on the temporary scene library collection A and scene coverage; the simulation scene is an OpenScenario scene library; S3, select the test vehicle in step S2 described in the test site of the test section, identify the scene in step S2 described in the temporary scene library in the test section, collect the scene data information, the scene data information is understood and classified; S4, obtain the single test time of the test vehicle in step S3 in the open road test section and the test route corresponding to the single test time, arrange and optimize the test route and test time according to the scene data information in step S3, and establish the optimal test section of intelligent driving car in open road.

2. The use of a system for open road testing of intelligent vehicles according to claim 1, characterized in that, The specific steps of classifying the scene information of the open road section in step S1 include: S11, select a test site for open road test of experimental vehicle to obtain scene information of test open road section; S12, according to the data provided by the road network information, the test section in step S11 is divided into special driving section and non special driving section; obtain the test road scene information included in the special driving section; S13, based on traffic big data and road network information collection scene information of test open road section in step S11; use the test road scene information included in the special driving section in step S12 to filter the open road scene information, and divide the scene information of the open road section into special driving section scene information and non special driving section scene information.

3. The method of claim 2, wherein, the temporary scene library collection in step S2 A The specific steps of the construction include: S21, input the special driving section scene information and non special driving section scene information in step S13 into the simulation scene for screening to obtain the screened scene; S22, building a temporary scene library set according to the scenes screened in step S21 A , building a test site based on the temporary scene library set A and the temporary scene coverage 4. The method of claim 3, wherein, The expression of the temporary scene coverage in step S22 is: wherein, c is the temporary scene coverage; Ω is the set of scenes of the complete scene library, A is the set of scenes of the temporary scene library, A = 1 - (1 - p)α .

5. The method of claim 3, wherein, The scene repetition number of the scene set A of the temporary scene library in step S22 is: in, R A collection of scenes for a temporary scene library A The number of scene repetitions; A i For scene collection A The first in i A temporary scene library, A j It is a collection of scenes A The Middle j A temporary scene library, i≠j When c=1, select the scene set from the temporary scene library. A The temporary scene library with the minimum number of scene repetitions R and its corresponding road segments are used as the final test road segments of the test site; where R≥0.

6. The method of claim 1, wherein, The specific steps of selecting the test vehicle in step S2 described in the test section of the test site in step S3 are as follows: S31, install laser radar and camera for the test vehicle; S32, test the open road section of the test vehicle driven by the test vehicle, repeat driving in the open road section, and collect the scene data information of the open road section. The scene of the open road section is the scene in the temporary scene library in step S2; S33, after the scene data information collection in step S32 is completed, the scene data information is understood and classified.

7. The method of claim 6, wherein the method further comprises: The specific steps of arranging and optimizing in step S4 include: S41, obtain the single test time of the test vehicle in step S32 in the open road test section and the test route corresponding to the single test time; S42, respectively input the scene classification data obtained in step S33 and the test route corresponding to the single test time in step S41 into the machine learning algorithm, analyze the time and space distribution relationship model of the specific scene type of the test route corresponding to the single test time by using the machine learning algorithm, and obtain the scene quantity of the test route corresponding to the single test time; S43, based on the single test time corresponding to the scene quantity and loss function formula of the test route of step S42, the optimal test section of the intelligent driving car on the open road is obtained.

8. The method of claim 6, wherein, The laser radar in step S31 is arranged on the roof of the experimental vehicle; the camera is arranged on the front and rear sections of the experimental vehicle.

9. The method of claim 6, wherein, The classification in step S33 is a static scene classification technology.

10. The method of claim 7, wherein, The loss function formula in step S43 is: wherein: x n*a represents the current point on the n-th special road longitude at each iteration, x n*b represents the next special road point selected on the n-th special road longitude at each iteration, y n*a represents the current point on the n-th special road latitude at each iteration, y n*b represents the next special road point selected on the n-th special road latitude at each iteration, num represents the number of scenes of the test route corresponding to a single test time.

Images