CN-121980277-A - Data processing method, electronic device, vehicle, medium, and program product

CN121980277ACN 121980277 ACN121980277 ACN 121980277ACN-121980277-A

Abstract

The application relates to the technical field of vehicle driving, in particular to a data processing method, electronic equipment, a vehicle, a medium and a program product, wherein the method comprises the steps of obtaining initial sample data of a first vehicle, wherein the first vehicle has a first driving direction; and training the model by utilizing the target sample data to obtain a perception model, wherein the perception model is suitable for a second vehicle, the second vehicle has a second driving direction, and the second driving direction is opposite to the first driving direction. Based on this, the difficulty and cost for training sample data to obtain a perception model corresponding to the second vehicle can be reduced.

Inventors

WU JIAXI
LI AISHI
GUAN JUN
LI XIANG

Assignees

智驾新程(上海)智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (16)

1. A data processing method applied to an electronic device, the method comprising: Obtaining initial sample data of a first vehicle, wherein the first vehicle has a first driving orientation; Mapping the initial sample data into target sample data; And training the model by using the target sample data to obtain a perception model, wherein the perception model is suitable for a second vehicle, the second vehicle has a second driving direction, and the second driving direction is opposite to the first driving direction.
2. The method of claim 1, wherein the first driving orientation is a left side of the vehicle body and the second driving orientation is a right side of the vehicle body, or wherein the first driving orientation is a right side of the vehicle body and the second driving orientation is a left side of the vehicle body.
3. The method according to claim 1, wherein the initial sample data includes first and second initial sample data having different data types, and The mapping the initial sample data to target sample data includes: mapping the first initial sample data by a first mapping mode, and Mapping the second initial sample data by adopting a second mapping mode; wherein the first mapping manner is different from the second mapping manner.
4. A method according to claim 3, wherein the initial sample data comprises at least one of: Image, camera internal parameters, camera external parameters, vehicle track data, moving obstacle track data, road structure marking data and point cloud data.
5. The method of claim 4, wherein the first initial sample data is an image, and wherein the first mapping means comprises: modifying a first pixel value of a first pixel in the image to a second pixel value of a second pixel, wherein the first pixel and the second pixel are symmetrical with respect to a longitudinal center axis of the image.
6. The method of claim 4, wherein the first initial sample data is a camera reference, and wherein the first mapping means comprises: And acquiring a first pixel coordinate value of a principal point in the camera internal parameter in an image coordinate system, and converting the first pixel coordinate value into a second pixel coordinate value, wherein a coordinate point corresponding to the first pixel coordinate value and a coordinate point corresponding to the second pixel coordinate value are symmetrical relative to a longitudinal central axis of the image.
7. The method of claim 4, wherein the first initial sample data is a camera overlay comprising a first transformation matrix of a first camera coordinate system and a vehicle coordinate system of the first vehicle, the first camera coordinate system corresponding to the first camera, and The first mapping mode comprises the following steps: And converting the first conversion matrix into a second conversion matrix, wherein the second conversion matrix is a conversion matrix of a second camera coordinate system and the vehicle coordinate system, and the positions and the postures of a second camera corresponding to the second camera coordinate system and the first camera are symmetrical relative to the longitudinal tangent plane of the first vehicle.
8. The method of claim 4, wherein the first initial sample data is the vehicle trajectory data, the vehicle trajectory data comprising at least one first vehicle pose matrix, the first vehicle pose matrix being a transformation matrix of a vehicle coordinate system and a first world coordinate system of the first vehicle; The first mapping mode comprises the step of converting the first vehicle pose matrix into a second vehicle pose matrix, wherein the second vehicle pose matrix is a conversion matrix of the vehicle coordinate system and a second world coordinate system, the second world coordinate system is obtained by right-hand chiral conversion of a middle world coordinate system, and the middle world coordinate system and the first world coordinate system are symmetrical relative to a longitudinal tangent plane of the first vehicle.
9. The method of claim 4, wherein the first initial sample data is the moving obstacle trajectory data, the moving obstacle trajectory data comprising at least one first trajectory point coordinate; The first mapping mode comprises the step of converting the first track point coordinate into a second track point coordinate, wherein the first track point coordinate and the second track point coordinate are symmetrical relative to the longitudinal tangent plane of the first vehicle.
10. The method of claim 4, wherein the first initial sample data is the road structure annotation data comprising first annotation coordinates of the road structure element and a category attribute of the road structure element, and The first mapping mode comprises the steps of converting the first labeling coordinate into a second labeling coordinate, wherein the first labeling coordinate and the second labeling coordinate are symmetrical relative to a longitudinal section of the first vehicle, and And modifying the category providing the direction semantics in the category attributes into opposite semantics.
11. The method of claim 4, wherein the first initial sample data is the point cloud data, and The first mapping mode comprises the following steps: Acquiring a first point cloud coordinate of the point cloud data in a radar coordinate system; Converting the first point cloud coordinates into second point cloud coordinates in a vehicle coordinate system of the first vehicle based on radar external parameters, wherein the radar external parameters are conversion matrixes of the radar coordinate system and the vehicle coordinate system; And converting the second point cloud coordinate into a third point cloud coordinate, wherein the second point cloud coordinate and the third point cloud coordinate are symmetrical relative to the longitudinal tangent plane of the first vehicle.
12. The method of any one of claims 1-11, wherein the perception model comprises an attention mechanism model.
13. An electronic device comprising a memory for storing instructions; at least one processor configured to execute the instructions to cause the electronic device to implement the data processing method of any one of claims 1-12.
14. A vehicle includes a memory for storing instructions; At least one processor configured to execute the instructions to cause the vehicle to implement the data processing method of any one of claims 1-12.
15. A computer readable storage medium having stored thereon instructions which, when executed on a computer, cause the computer to perform the data processing method of any of claims 1-12.
16. A computer program product, characterized in that the computer program product, when run on a device, causes the device to perform the data processing method of any of claims 1-12.

Description

Data processing method, electronic device, vehicle, medium, and program product Technical Field The present application relates to the field of vehicle driving technologies, and in particular, to a data processing method, an electronic device, a vehicle, a medium, and a program product. Background The traffic system is different in different areas. The traffic system in regions such as korea, thailand, malaysia and the like is constructed according to left rudder right driving rules, that is, the driving direction of the vehicle is on the left side of the vehicle body and is on the right side of the road. Traffic systems in the regions of the united kingdom, australia, india, japan, hong kong and the like are all constructed according to the right rudder left driving rule, namely, the vehicle driving direction is on the right side of the vehicle body and is driven on the left side of the road. It will be appreciated that the performance of the perceptual model needs to depend on the amount of sample data used for training. After a large amount of sample data is collected in the left rudder scene, the perception model is trained based on the sample data of the large amount of left rudder scene, and the obtained perception model in the left rudder scene can achieve higher performance. And if the perceived model in the right rudder scene is required to achieve higher performance, sample data in the right rudder scene is required to be acquired as well. However, because the traffic systems of different areas are different, the cost for acquiring the sample data under a sufficient number of right rudder scenes is high, and the difficulty is high. Disclosure of Invention The embodiment of the application provides a data processing method, which solves the problem of high sample data acquisition cost for training to obtain a required perception model. In a first aspect, an embodiment of the present application provides a data processing method, applied to an electronic device, where the method includes obtaining initial sample data of a first vehicle, where the first vehicle has a first driving direction, mapping the initial sample data to target sample data, and performing model training with the target sample data to obtain a perception model, where the perception model is applicable to a second vehicle, where the second vehicle has a second driving direction, and the second driving direction is opposite to the first driving direction. It can be appreciated that the initial sample data of the first vehicle with the first driving direction is mapped into the target sample data, and the perception model suitable for the second vehicle can be trained by using the target sample data, so that the sample data volume available for training to obtain the perception model suitable for the second vehicle is greatly increased, and the difficulty and cost for acquiring the sample data for training to obtain the perception model corresponding to the second vehicle are reduced. After training is performed by using the target sample data obtained after mapping and a perception model is obtained, the perception model obtained after training can be ensured to reach the required performance. In one possible implementation of the first aspect, the first driving direction is a left side of the vehicle body, the second driving direction is a right side of the vehicle body, or the first driving direction is a right side of the vehicle body, and the second driving direction is a left side of the vehicle body. It can be appreciated that when the first driving direction is the left side of the vehicle body, the second driving direction is the right side of the vehicle body, that is, the initial sample data is the sample data in the left rudder scene, and the target sample data is the sample data for training to obtain the perception model corresponding to the right rudder scene. When the first driving direction is the right side of the vehicle body, the second driving direction is the left side of the vehicle body, namely, the initial sample data are sample data in a right rudder scene, and the target sample data are sample data for training to obtain a perception model corresponding to the left rudder scene. In one possible implementation manner of the first aspect, the initial sample data includes first initial sample data and second initial sample data with different data types, and mapping the initial sample data into the target sample data includes mapping the first initial sample data in a first mapping manner and mapping the second initial sample data in a second mapping manner, where the first mapping manner is different from the second mapping manner. In a possible implementation of the first aspect, the initial sample data includes at least one of an image, a camera internal parameter, a camera external parameter, vehicle track data, moving obstacle track data, road structure marking data, and point cloud data. In one possible