Multiple view point cloud registration based on 3D lines

A point cloud registration method based on 3D lines extraction from 3D data to register point cloud with obvious edges is proposed in this paper. Firstly, the line feature point cloud (LFPC), which is corresponding to the objects' edges, is extracted from the measured 3D data by using surface curvature as a measure. Then, through applying the 3D Hough transformation on LFPC, the line directions and positions are extracted respectively. Thus, the 3D lines corresponding to the edges in the two point clouds to be registered are obtained. Finally, a 3D lines matching and registration algorithm is proposed to accomplish the registration. Furthermore, the experiment is conducted to testify the feasibility of proposed method. The contribution of this paper is to propose a novel 3D Hough based lines extraction algorithm as well as a novel 3D line matching and registration algorithm, which can also be used in other 3D point cloud processing.