Investigation into registration of scanned 3D image based on geometric feature identification

Abstract

This paper investigates geometric feature extraction from scanned image and applies it in multi-view image registration. The presented registration approach includes three steps, feature extraction, coarse registration and fine registration. Firstly, feature points are identified based on curvature estimation, and feature point linkage is set up according to neighboring relationship of the extracted feature points. The coarse registration is conducted by alignment transmission calculation using the overlapping feature linkages extracted from the two-view images. Finally, iterative Closest Point (ICP) is used in fine registration. Experimental results of multi-view images taken by laser scanner are carried out to compare the convergence and registration error between the presented approaches with classical ICP. The presented registration approach achieves higher convergence than classical ICP, and can overcome the problems of traditional ICP in low overlapping and bad initial estimate.