Point cloud processing method and system based on random downsampling

Abstract

The invention discloses a cloud processing method and system based on random reduced sampling points. The method comprises the following steps: carrying out local feature aggregation on an original point cloud; carrying out random downsampling on the point cloud after local feature aggregation; and processing the sparse point cloud after random downsampling. According to the method, simple and efficient random downsampling is adopted to greatly reduce the number of point clouds, and meanwhile, a specially designed feature aggregation strategy is adopted to ensure the retention of effective information, so that the spatial information and geometric information of the point clouds can be retained to the greatest extent under the condition of obviously reducing the calculation amount and memory occupation, and a more efficient and accurate processing result can be obtained.

Description

technical field [0001] The invention belongs to the field of point cloud processing, and in particular relates to a point cloud processing method based on random downsampling, an encoder-decoder structure, a system, electronic equipment and a storage medium. Background technique [0002] Although the RGB image can capture the apparent texture of the object well, it cannot intuitively reflect the spatial information of the object. 3D point cloud data can provide object space information, which can make up for the weakness of RGB image in capturing space information. Therefore, 3D point cloud data has been more and more widely used in the field of computer vision. However, point cloud data often contains hundreds of millions of point information, which consumes a lot of computing resources and is not efficient. Commonly used point cloud processing methods are: processing based on original points, processing based on projection, and processing based on voxelization. Based on...

AI Technical Summary

Problems solved by technology

Based on the original point processing method, the original point cloud will be directly manipulated, and the spatial information can be preserved to the maximum extent. If all point clouds are calculated at one time, the computing resources will be very expensive, so the original point cloud will generally be divided into smaller point cloud blocks and then processed. processing, but this will destroy the geometric information of the object; the projection-based processing method converts the 3D point cloud to a 2D plane through spherical projection and other methods for processing, although the processing process is simplified, but the spatial information is greatly sacrificed; The voxelization-based processing divides the point cloud into small grids, and uses one point to represent all the points in the grid, which reduces the computational complexity while retaining certain spatial information, but when the number of point clouds is large The effect of reducing computational complexity is not obvious, and when the point cloud density is uneven, dense grids tend to lose more information, while dense grids usually contain important information

Images