Rapid registration method by using partial 3D point clouds

Abstract

Due to the rapid development of three-dimensional laser technique, the registration algorithm attracts more attention in industrial application. In this paper, a rapid registration method of 3D point clouds is proposed to improve the efficiency of reverse engineering. The method is based on weighted principal analysis method and re-weighted iterative closest point algorithm. The pipeline consists three stages: coarse alignment, data simplification, and fine alignment. During the first phase, a weighted principal component analysis method is used to calculate original transformation matrices between model data and scene data. In the second part, principal component eigenvalue analysis scheme is implemented to filter the redundant points of datasets. This procedure could reduce the input number of fine registration and refine outliers. Finally, the fine registration is obtained by a weighted iterative closest point algorithm based on Cauchy's and Welsch's function using the partial points obtained in the second phase. The experimental results and comparison experiments are utilized to demonstrate the validity and effectiveness of the proposed matching algorithm.