Markerless Point Cloud Matching Algorithm Based on 3D Feature Extraction

Abstract

Since 3D scanners can only scan in a limited range and the scanning process is prone to occlusion and other problems, a complete 3D model cannot be obtained from a point cloud scanning result. Therefore, point cloud matching is necessary for most 3D scanning projects. We propose a point cloud registration algorithm based on a comprehensive approach, and the research contents include a new curvature-based point cloud feature extraction method, a three-dimensional spatial structure classifier of feature points, and an unmarked point cloud-matching algorithm based on three-dimensional feature extraction. The experiments are based on the simulation data and real data to verify the algorithm and evaluate the accuracy. The experimental results show that the matching accuracy reaches the millimetre level, and the fully automated and high-precision label-free point cloud matching based on 3D features is realized, which can provide innovative and breakthrough help for 3D reconstruction.