CN-121999165-A - Road height prediction method, vehicle and storage medium

CN121999165ACN 121999165 ACN121999165 ACN 121999165ACN-121999165-A

Abstract

The embodiment of the application provides a road height prediction method, a vehicle and a storage medium, wherein the method comprises the steps of acquiring a road image of a road where a vehicle is located, a point cloud image of the road and a vehicle position of the vehicle in the running process of the vehicle; the method comprises the steps of generating a bird's-eye view angle grid map based on a vehicle position and preset regional parameters, carrying out attention enhancement processing on a road image based on the bird's-eye view angle grid map to obtain image bird's-eye view angle characteristics, carrying out structural feature extraction on a point cloud image to obtain point cloud bird's-eye view angle characteristics, and carrying out feature fusion on the image bird's-eye view angle characteristics and the point cloud bird's-eye view angle characteristics to predict and obtain the target height of a road. The method solves the technical problem of low road height prediction precision.

Inventors

GAO YONGJI
LU YUEJIE
CHEN ZHONGQIU

Assignees

安徽开阳科技有限公司
奇瑞汽车股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. A method of predicting the height of a road, comprising: Acquiring a road image of a road where a vehicle is located, a point cloud image of the road and a vehicle position of the vehicle in the running process of the vehicle; Generating an aerial view angle grid map based on the vehicle position and a preset area parameter, wherein different grid cells in the aerial view angle grid map correspond to query vectors of different aerial views in the road; Performing attention enhancement processing on the road image based on the bird's-eye view grid map to obtain an image bird's-eye view characteristic, wherein the attention enhancement processing is used for enhancing the characteristic associated with the road structure of the road in the road image; Carrying out structural feature extraction on the point cloud image to obtain a point cloud aerial view angle feature; And carrying out feature fusion on the image aerial view angle features and the point cloud aerial view angle features to predict and obtain the target height of the road, wherein the target height is used for representing the height of the road relative to a preset reference height.
2. The method according to claim 1, wherein performing attention enhancement processing on the road image based on the bird's-eye view grid map to obtain an image bird's-eye view feature, comprises: Predicting the height of the road based on the bird's eye view grid map to obtain the initial height of the road and the initial height confidence of the initial height; Extracting image features of the road image to obtain initial image features; And determining the image aerial view feature based on the initial height, the initial height confidence, the initial image feature and the aerial view grid map.
3. The method of claim 2, wherein determining the image aerial view feature based on the initial height, the initial height confidence, the initial image feature, and the aerial view grid map comprises: performing coordinate conversion on the initial height based on a preset coordinate system to obtain a three-dimensional position coordinate; Performing feature coding on the three-dimensional position coordinates to obtain position coding features; And carrying out feature aggregation on the initial image features based on the position coding features, the initial height confidence and the aerial view perspective grid map to obtain the image aerial view perspective features.
4. The method of claim 2, wherein predicting the elevation of the road based on the bird's eye view grid map results in an initial elevation of the road and an initial elevation confidence of the initial elevation, comprising: If the output image aerial view angle characteristics exist, carrying out height prediction on the road based on the output image aerial view angle characteristics and the aerial view angle grid map to obtain the initial height and the initial height confidence; And if the output image aerial view angle characteristics do not exist, carrying out height prediction on the road based on the aerial view angle grid map, and obtaining the initial height and the initial height confidence.
5. The method of claim 1, wherein feature fusion of the image aerial view feature and the point cloud aerial view feature to predict a target height of the road comprises: Performing space fusion on the aerial view angle characteristics of the image and the aerial view angle characteristics of the point cloud to obtain a first fusion result; Performing cross attention processing on the image aerial view angle characteristics and the point cloud aerial view angle characteristics to obtain a second fusion result; splicing the first fusion result and the second fusion result to obtain a fusion result; And predicting the target height based on the fusion result.
6. The method of claim 5, wherein spatially fusing the image aerial view feature and the point cloud aerial view feature to obtain a first fused result comprises: Determining image geometric information of the road image and point cloud geometric information of the point cloud image, wherein the image geometric information is used for reflecting structural characteristics of a road surface in the road image, and the point cloud geometric information is used for reflecting spatial characteristics of the road surface in the point cloud image; confidence prediction is carried out on the image geometric information and the point cloud geometric information respectively, so that image geometric information weight of the image geometric information and point cloud geometric information weight of the point cloud geometric information are obtained; And performing space fusion on the image aerial view angle characteristic, the image geometric information weight, the point cloud aerial view angle characteristic, the point cloud geometric information and the point cloud geometric information weight to obtain the first fusion result.
7. The method of claim 1, wherein the performing structural feature extraction on the point cloud image to obtain a point cloud aerial view feature comprises: Carrying out space division on the point cloud image to obtain a plurality of voxel features, wherein different voxel features are used for recording area information of different areas corresponding to the point cloud image; performing feature aggregation on the plurality of voxel features based on the geometric information of the point cloud image to obtain structural voxel features; and performing dimension conversion on the structural voxel characteristics to obtain the point cloud aerial view angle characteristics.
8. The method of claim 1, wherein generating a bird's eye view gridding based on the vehicle location and a preset zone parameter comprises: Creating an initial aerial view angle grid map based on the preset area parameter by taking the vehicle position as the center, wherein the initial aerial view angle grid map comprises a plurality of grid cells, and aerial view angle information of different aerial views in the road is stored in different grid cells; And updating the initial aerial view grid map based on the aerial view information of the different aerial views to obtain the aerial view grid map.
9. The method of claim 8, wherein updating the initial bird's-eye view angle grid map based on the bird's-eye view angle information of the different bird's-eye views to obtain the bird's-eye view angle grid map comprises: mapping the bird's-eye view angle information of the different bird's-eye view angles into different grids in the initial bird's-eye view angle grid map to obtain a mapping result; and fusing the mapping result and the aerial view angle information corresponding to the different grids to obtain the aerial view angle grid map.
10. A vehicle, characterized by comprising: A memory storing an executable program; a processor for executing the program, wherein the program when run performs the method of any of claims 1 to 9.

Description

Road height prediction method, vehicle and storage medium Technical Field The embodiment of the application relates to the technical field of automatic driving, in particular to a road height prediction method, a vehicle and a storage medium. Background In the field of autopilot, related art generally relies on monocular vision or lidar data for road surface elevation prediction. However, when facing complex road surfaces, such as geometrically discontinuous areas like slopes and depressions, the lack of depth semantic information makes the vehicle's understanding of the road conditions less accurate, resulting in a lower accuracy of the road surface height predictions. There is currently no good solution to the above problems. Disclosure of Invention The embodiment of the application provides a road height prediction method, a vehicle and a storage medium, which are used for at least solving the technical problem of low road surface height prediction precision. According to one aspect of the embodiment of the application, a road height prediction method is provided, which comprises the steps of obtaining a road image of a road where a vehicle is located, a point cloud image of the road and a vehicle position of the vehicle in the process of driving the vehicle, generating a bird's-eye view grid map based on the vehicle position and preset area parameters, wherein different grid units in the bird's-eye view grid map correspond to query vectors of different bird's-eye views in the road, performing attention enhancement processing on the road image based on the bird's-eye view grid map to obtain an image bird's-eye view feature, wherein the attention enhancement processing is used for performing enhancement processing on features related to a road structure of the road in the road image, performing structural feature extraction on the point cloud image to obtain the point cloud bird's-eye view feature, and performing feature fusion on the image bird's-eye view feature and the point cloud bird's-eye view feature to predict and obtain a target height of the road, wherein the target height is used for representing the height of the road relative to a preset reference height. Further, attention enhancement processing is conducted on the road image based on the aerial view grid map to obtain aerial view characteristics of the image, the method comprises the steps of conducting height prediction on the road based on the aerial view grid map to obtain initial height of the road and initial height confidence of the initial height, conducting image characteristic extraction on the road image to obtain initial image characteristics, and determining the aerial view characteristics of the image based on the initial height, the initial height confidence, the initial image characteristics and the aerial view grid map. Further, the method comprises the steps of determining the aerial view angle characteristics of the image based on the initial height, the initial height confidence, the initial image characteristics and the aerial view angle grid map, carrying out coordinate conversion on the initial height based on a preset coordinate system to obtain three-dimensional position coordinates, carrying out feature coding on the three-dimensional position coordinates to obtain position coding characteristics, and carrying out feature aggregation on the initial image characteristics based on the position coding characteristics, the initial height confidence and the aerial view angle grid map to obtain the aerial view angle characteristics of the image. The method comprises the steps of obtaining initial height and initial height confidence of a road, wherein the initial height and the initial height confidence of the road are obtained by carrying out height prediction on the road based on the output image aerial view angle characteristic and the aerial view angle grid diagram if the output image aerial view angle characteristic exists, obtaining the initial height and the initial height confidence, and carrying out height prediction on the road based on the aerial view angle grid diagram if the output image aerial view angle characteristic does not exist, and obtaining the initial height and the initial height confidence. Further, feature fusion is conducted on the image aerial view angle features and the point cloud aerial view angle features to predict the target height of the road, and the method comprises the steps of conducting space fusion on the image aerial view angle features and the point cloud aerial view angle features to obtain a first fusion result, conducting cross attention processing on the image aerial view angle features and the point cloud aerial view angle features to obtain a second fusion result, conducting splicing on the first fusion result and the second fusion result to obtain a fusion result, and predicting the target height based on the fusion result. Further, spatial fusion is carried out on the ima