Voxel processing method, apparatus, storage device, and electronic medium

By uniformly dividing the voxels of a 3D point cloud into sets and extracting the set features, the problem of low efficiency of neural networks in 3D point cloud prediction tasks is solved, and more efficient prediction task processing is achieved.

CN117237731BActive Publication Date: 2026-07-03北京亮道智能汽车技术有限公司
1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
北京亮道智能汽车技术有限公司
Filing Date
2023-09-22
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

In existing technologies, when using neural networks to complete prediction tasks based on multiple voxels in 3D point clouds, there is a problem of low task processing efficiency, mainly due to the uneven distribution of voxels in 3D point clouds, which leads to high data complexity of neural networks.

Method used

The N voxels corresponding to the 3D point cloud are divided into M voxel sets, each containing the same number of voxels. By extracting the set features of each voxel set and inputting them into the neural network, the computational complexity of the neural network is reduced.

Benefits of technology

This method achieves uniform division of 3D point clouds, reduces the complexity of neural networks in determining the category information of each point, and improves the processing efficiency of prediction tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117237731B_ABST
    Figure CN117237731B_ABST
Patent Text Reader

Abstract

This application discloses a voxel processing method, apparatus, storage medium, and electronic device. The method includes: dividing N voxels into M voxel sets based on the positional feature information of each voxel in an N voxel set corresponding to a 3D point cloud, wherein each voxel set contains the same number of voxels; extracting set features corresponding to each voxel set in the M voxel sets, wherein the set features corresponding to each voxel set include the correlation relationships between voxels within that voxel set; and inputting the set features corresponding to each voxel set into a neural network to obtain the category information of each point in the 3D point cloud determined by the neural network based on the set features corresponding to each voxel set. This application solves the technical problem of low task processing efficiency in existing technologies when using neural networks to complete prediction tasks based on multiple voxels in a 3D point cloud.
Need to check novelty before this filing date? Find Prior Art