A Study on a Planar Point Cloud ICP Registration Algorithm Based on Boundary Extraction

The demand for 3D scanning of workpiece geometries in automated assembly within workshops is increasingly critical, playing a vital role in the process. Point cloud registration, as an essential component of 3D data processing, has become both highly demanded and significant. However, in large-scale planar 3D point cloud registration, challenges such as excessive data volume and feature point mismatches often lead to prolonged registration times and unstable accuracy. To deal with these issues, this paper proposes a planar point cloud ICP registration algorithm based on boundary extraction. The algorithm optimizes the registration process by incorporating boundary extraction, keypoint extraction through the normal vector angle, SHOT descriptors, and RANSAC-based coarse registration. This approach effectively mitigates inaccuracies in outlier removal and feature point mismatches in boundary point clouds, reducing the distance between the target and source point clouds. The Iterative Closest Point (ICP) algorithm is then applied for precise registration using the boundary point cloud. Experimental results show that this approach enhances boundary extraction efficiency by 50.6912% compared to the SOR extraction method. Additionally, when compared to the ICP, FPFH+ICP, and NDT algorithms, the proposed method achieves an average registration speed increase of 95.527%, 61.449%, and 22.69%, respectively, meeting the demands of automated assembly.