Search

CN-122020352-A - Intelligent driving data generation method and system based on real vehicle test

CN122020352ACN 122020352 ACN122020352 ACN 122020352ACN-122020352-A

Abstract

The invention relates to an intelligent driving data generation method based on real vehicle testing, which comprises the steps of constructing a scene event library, screening vehicle data of scene events affecting intelligent driving safety from real vehicle testing data based on the scene event library, selecting event information elements of the scene events, converting the vehicle data into a format containing each event information element, classifying and outputting the vehicle data based on the event information elements, and changing real vehicle testing data into active generation from simple passive records through a three-level architecture of real-time screening at a vehicle-mounted end, edge data calculation pretreatment and cloud closed-loop generation, so that efficient extraction, rapid processing and closed-loop utilization of high-value scene events are realized, and a low-cost, high-efficiency and sustainable testing data engine is provided for an intelligent driving system.

Inventors

  • ZHANG QIFENG

Assignees

  • 武汉江夏楚能汽车技术研发有限公司

Dates

Publication Date
20260512
Application Date
20251212

Claims (10)

  1. 1. The intelligent driving data generation method based on the real vehicle test is characterized by comprising the following steps of: Step 1, constructing a scene event library, wherein the scene event library records scene data of each scene event affecting intelligent driving safety; Step 2, screening from real vehicle test data based on the scene data of the scene event library to obtain vehicle data of scene events affecting intelligent driving safety, wherein the vehicle data are data acquired by a sensor and a processor in the running process of the vehicle; Step 3, selecting event information elements of scene events, and converting the vehicle data into a format containing each event information element; and 4, classifying the vehicle data based on the event information elements and outputting the classified vehicle data.
  2. 2. The intelligent driving data generating method according to claim 1, wherein the acquiring procedure of the scene data of any one of the scene events in the scene event library in step 1 includes: And acquiring a plurality of scene data of any scene event affecting intelligent driving safety to construct a training set, inputting the training set into an AI model identified by the scene event, and outputting the scene data of the corresponding scene event by the AI model identified by the scene event.
  3. 3. The intelligent driving data generating method according to claim 1, wherein scene events in the scene event library are pruned or added in real time according to requirements; The scene events affecting intelligent driving safety comprise collision risk events, lane departure events, solid lane change events, ‌ deceleration exceeding a safety threshold event, ‌ steering wheel angle abnormality ‌ events and system and environment type alarm events; the scene data of the collision risk event includes that a distance to an active human or vehicle is less than a safety threshold; the scene data of the lane departure event comprises that the vehicle deviates from the current lane under the condition that a turn signal is not turned on; the scene data of the solid line lane change event comprises lane change of a vehicle in a solid line area of lane change prohibition; The ‌ deceleration exceeding the safety threshold event includes the vehicle deceleration being greater than a threshold; The scene data of the ‌ steering wheel angle anomaly ‌ event comprises that the steering wheel change speed is greater than a threshold value, the steering wheel change angle is greater than the threshold value or the steering wheel angle is alternately changed frequently; The system and environment class alert event scenario data includes a sensor failure or a perceived capability of the vehicle being below a threshold.
  4. 4. The intelligent driving data generating method according to claim 1, wherein the step 2 further comprises setting a time range, and collecting vehicle data of the set time range before and after the occurrence of the scene event.
  5. 5. The intelligent driving data generating method according to claim 1, wherein the step 3 further comprises deploying edge computing nodes at a test field or fleet management station; The edge computing node receives the time data uploaded by the vehicle data, aligns various data in the vehicle data according to time stamps, performs classification marking on categories and behaviors of targets in videos/images acquired by the sensor by utilizing a pre-training model, converts the vehicle data into a format containing various event information elements, and then guides the format into a cloud database.
  6. 6. The intelligent driving data generating method according to claim 5, wherein the vehicle data includes data acquired by a camera, a laser radar and an IMU; classifying the object in the video/image comprises vehicles, pedestrians and signal lamps, and classifying and labeling the behavior of the object in the video/image comprises turning and crossing; The event information elements include scene type, time stamp, location, participation object and key parameters.
  7. 7. The intelligent driving data generating method according to claim 1, wherein the step 4 comprises classifying the vehicle data based on the event information element by a cloud database to generate: Simulation test cases for algorithm regression testing; according to the scene geographic position, automatically scheduling a real vehicle retest case for testing the fact that the vehicle goes to the same road section to carry out scene retest; Based on the existing scene, a derivative scene is generated through parameter disturbance, and the test cases of the data enhancement template of the test coverage are expanded.
  8. 8. The intelligent driving data generation system based on the real vehicle test is characterized by comprising a vehicle-mounted end real-time scene identification and data screening module, an edge data preprocessing module and a cloud; The vehicle-mounted terminal real-time scene recognition and data screening module is used for constructing a scene event library, and the scene event library records scene data of each scene event affecting intelligent driving safety; screening from real vehicle test data based on the scene data of the scene event library to obtain vehicle data of scene events affecting intelligent driving safety, wherein the vehicle data are data acquired by a sensor and a processor in the running process of the vehicle; The edge data preprocessing module is used for selecting event information elements of scene events and converting the vehicle data into a format containing each event information element; the cloud end is used for classifying and outputting the vehicle data based on the event information elements.
  9. 9. An electronic device comprising a memory, a processor for implementing the steps of the intelligent driving data generating method based on real vehicle testing as claimed in any one of claims 1-7 when executing a computer management class program stored in the memory.
  10. 10. A computer-readable storage medium, having stored thereon a computer-management-class program which, when executed by a processor, implements the steps of the intelligent driving data generation method based on real vehicle testing as claimed in any one of claims 1 to 7.

Description

Intelligent driving data generation method and system based on real vehicle test Technical Field The invention relates to the technical field of intelligent driving, in particular to an intelligent driving data generation method based on real vehicle testing. Background At present, in the development process of an intelligent driving system, a real vehicle test can generate massive raw data (such as a sensor, a vehicle state, a surrounding environment state, a driver operation and the like), but the prior art has some defects: 1. The data utilization rate is low, more than 90% of original data cases belong to invalid cases in real vehicle tests, such as regular scenes such as uniform speed, straight running and the like, and high-value edge scenes cannot be covered, so that the data marking cost is high, and resources are wasted. 2. The processing efficiency is poor, the traditional data processing method needs to store all original data firstly, then the data is cleaned, marked and modeled off-line, the single test data processing period can be up to several days, and the version iteration speed is seriously dragged. 3. The test closed loop is missing, namely data acquired by a real vehicle is difficult to be quickly converted into reproducible test cases, so that the research and development period of test-finding problem-reproduction problem is prolonged, and the test efficiency is low. Disclosure of Invention Aiming at the technical problems existing in the prior art, the invention provides an intelligent driving data generation method and system based on real vehicle testing, which changes real vehicle testing data from simple passive record to active generation of testing asset through a three-level architecture of 'real-time screening of a vehicle-mounted end + edge data calculation pretreatment + cloud closed-loop generation', realizes efficient extraction, rapid processing and closed-loop utilization of high-value scene events, and provides a low-cost, high-efficiency and sustainable testing data engine for the intelligent driving system. According to a first aspect of the present invention, there is provided an intelligent driving data generating method based on real vehicle testing, including: Step 1, constructing a scene event library, wherein the scene event library records scene data of each scene event affecting intelligent driving safety; Step 2, screening from real vehicle test data based on the scene data of the scene event library to obtain vehicle data of scene events affecting intelligent driving safety, wherein the vehicle data are data acquired by a sensor and a processor in the running process of the vehicle; Step 3, selecting event information elements of scene events, and converting the vehicle data into a format containing each event information element; and 4, classifying the vehicle data based on the event information elements and outputting the classified vehicle data. On the basis of the technical scheme, the invention can also make the following improvements. Optionally, the acquiring process of the scene data of any scene event in the scene event library in step 1 includes: And acquiring a plurality of scene data of any scene event affecting intelligent driving safety to construct a training set, inputting the training set into an AI model identified by the scene event, and outputting the scene data of the corresponding scene event by the AI model identified by the scene event. Optionally, the scene events in the scene event library are pruned or added in real time according to the requirements; The scene events affecting intelligent driving safety comprise collision risk events, lane departure events, solid lane change events, ‌ deceleration exceeding a safety threshold event, ‌ steering wheel angle abnormality ‌ events and system and environment type alarm events; the scene data of the collision risk event includes that a distance to an active human or vehicle is less than a safety threshold; the scene data of the lane departure event comprises that the vehicle deviates from the current lane under the condition that a turn signal is not turned on; the scene data of the solid line lane change event comprises lane change of a vehicle in a solid line area of lane change prohibition; The ‌ deceleration exceeding the safety threshold event includes the vehicle deceleration being greater than a threshold; The scene data of the ‌ steering wheel angle anomaly ‌ event comprises that the steering wheel change speed is greater than a threshold value, the steering wheel change angle is greater than the threshold value or the steering wheel angle is alternately changed frequently; The system and environment class alert event scenario data includes a sensor failure or a perceived capability of the vehicle being below a threshold. Optionally, the step 2 further includes setting a time range, and collecting vehicle data of the set time range before and after the occurrence time of the sce