A point cloud segmentation algorithm using local convexity and octree

An algorithm of point cloud segmentation using octree and local convexity is proposed to deal with the fact that normal segmentation methods have poor effect on coarse point clouds. The proposed algorithm is composed of three steps. Firstly, point clouds with 3D coordinates are obtained using special instruments. Then, normal estimation approaches are used to get the normal vector and to segment the point clouds into patches by octree. Finally, the neighboring patches are merged if local convexity is held. Compared with other works, the proposed algorithm can reduce the number of surface effectively and produce surfaces with better quality. Experimental results show that the proposed algorithm can better segment the nearly and equally distributed data into several meaningful surfaces and 90% of the surfaces is the same as that produced by human operation.