Point cloud registration for a non-deformable object using surface curvature features

Recently, many applications require 3D information of an object (i.e. point cloud) such as 3D image scanning, 3D surface reconstruction and etc. However many researches on point cloud registration face many challenges especially for increasing the accuracy of point cloud matching. This research employs surface curvature features in discrete surfaces. A surface curvature feature is a pointer of ridges that are invariant to rigid body transformations. In this paper, a new algorithm of point cloud registration for non deformable object is proposed. This algorithm employs surface curvature features estimated by fitting knearest neighbor of local point to hyperbolic paraboloid equation. The proposed algorithm is implemented with Iterative Closest Point (ICP) technique and quantitatively evaluated and compared with common techniques for point cloud registration. Experimental results demonstrate that the proposed technique approximately 63% faster and 23% more accurate than iterative closest point with angular invariant feature (ICP-AIF) registration techniques. These results are obtained by testing the proposed frame work with noise, different pose of point cloud and overlapped area.