Fully automatic registration of structured objects based on laser radar range images

Abstract

3D data registration is important in object modeling, and it is usually performed in two steps: coarse registration and fine registration. In order to automatically align laser radar (Ladar) range images of structured objects and avoid any feature detection or manual rough alignment, a 3DPEICP (iterative closest point based on object 3D pose estimation) registration method is proposed. Focusing on the coarse registration, the transformation parameters required for roughly aligning the 3D datasets are acquired by the 3D pose mainly estimated with the normals of large horizontal top surfaces and vertical sides of the targets. Fine registration is completed through the Iterative Closest Point (ICP) method within a few iterations for improving alignment accuracy. Experimental results of different models demonstrate the validity of the approach.