A method of regular objects recognition from 3D laser point cloud

Abstract

Outside and inside scenes are usually made of regular surfaces such as plane, spherical surface, cylindrical surface, and so on. While we capture 3D data of the real scenes using the laser scanner, 3D laser point cloud includes a lot of, regular surfaces. So we presented a method recognizes the plane, spherical and cylindrical surface based on normal and curvature. We used Coordinate Transformation method (CT) to solve normal and curvature; used linear octree improve the computing efficiency of general CT method. We gave the experiment for testing the robustness of the algorithm. © Springer-Verlag Berlin Heidelberg 2012.