Registration of Scattered Cloud Data

An algorithm to automatically register point cloud data from different scans is proposed.The source data may include no additional information no other than coordinates of the measured points.In the algorithm,normal vector and curvature are first calculated according to the point and its neighbor points, and then normal vector of every point is oriented to the outside of the surface.All the pair wise points are located whose curvatures are sufficiently si milar and all the rigid transformations that map the first point to the second one are computed, while making the normal vectors coincide.A hash table is constituted from the coordinate transformations in 3D space.The target transformation that makes the most amount normal vectors coincide in this table is employed to register the two point cloud data.Thus, initialized by the former result, the iterative closest point algorithm leads to perfect registration.Experimental results show the accurate and robust performance of the proposed algorithm.