Robust Simultaneous Registration of Multiple Range Images Comprising a Large Number of Points

摘要

As part of a project for the electronic preservation of cultural objects, this paper proposes a new technique for position alignment of multiple range images. In particular, the study focuses on accuracy of alignment and robustness to noise in the input data. Using the conjugate gradient method based on the M-estimator, simultaneous alignment is performed by minimization of an error function that handles all range images at the same time. Furthermore, for more efficient and more accurate search for matching points, the KD-tree data structure is utilized, and additional attributes such as the laser reflection brightness and the brightness of the three-dimensional point itself are introduced into the error measure. This paper proposes such a method and presents the results of experiments with real data.