Farthest point sampling acceleration method, system, device, medium and program product for three-dimensional point cloud

By optimizing point cloud sampling through multi-stage processing and modular hardware architecture, the problem of high complexity in large-scale point cloud FPS algorithms is solved, achieving low latency and high throughput acceleration of point cloud sampling.

CN122336151APending Publication Date: 2026-07-03SHANGHAI TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHANGHAI TECH UNIV
Filing Date
2026-05-18
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies suffer from high computational complexity in processing large-scale point clouds due to the FPS algorithm and the lack of hardware acceleration solutions optimized for large-scale point clouds, making it difficult to meet real-time processing requirements.

Method used

A multi-stage processing approach is adopted, including coarse sampling, global distance update and fine sampling stages, combined with a modular hardware architecture and a neighborhood voxel selector to optimize the sampling process of point clouds.

Benefits of technology

It achieves low-latency, high-throughput sampling of farthest points for large-scale point clouds, accelerates FPS operations, reduces memory access overhead and hardware resource redundancy, and adapts to the real-time processing needs of point clouds of different sizes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This invention provides a method, system, device, medium, and program product for accelerating farthest point sampling of 3D point clouds, relating to the fields of 3D point cloud processing and hardware acceleration. The method includes: voxelization preprocessing of the point cloud, dividing the 3D space of the point cloud into multiple voxel grids, with each point categorized into a corresponding voxel based on its spatial coordinates; and performing multi-stage processing on the point cloud, including a coarse sampling stage, a global distance update stage, and a fine sampling stage. This invention is the first to decompose FPS into three stages: "coarse sampling - distance update - fine sampling," with an adaptive switching strategy based on point cloud density. Unlike existing methods that still require accessing the entire point cloud in the early stages, this invention utilizes voxel centers to replace the point cloud for coarse-grained sampling, fundamentally solving the bottleneck problem of high memory access overhead in the early sampling stage of large-scale point clouds.
Need to check novelty before this filing date? Find Prior Art