Robust matching of 3D contours using iterative closest point algorithm improved by M-estimation

摘要

An extension of the iterative closest point matching by M-estimation is proposed for realization of robustness to non-overlapping data or outlying data in two sets of contour data or depth images for rigid bodies. An objective function which includes independent residual components for each of x, y and z coordinates is originally defined and proposed to evaluate the fitness, simultaneously dealing with a distribution of outlying gross noise. The proposed procedure is based on modified M-estimation iterations with bi-weighting coefficients for selecting corresponding points for optimization of estimating the transforms for matching. The transforms can be represented by 'quaternions' in the procedure to eliminate redundancy in representation of rotational degree of freedom by linear matrices. Optimization steps are performed by the simplex method because it does not need computation of differentiation. Some fundamental experiments utilizing real data of 2D and 3D measurement show effectiveness of the proposed method. When reasonable initial positions are given, the unique solution of position could be provided in spite of surplus point data in the objects. And then the outlying data could be filtered out from the normal ones by the proposed method. (C) 2003 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. [References: 13]