Automatic multi-view registration of unordered range scans without feature extraction

摘要

For 3D object modeling, this paper proposes an automatic approach for multi-view registration of unordered range scans by both the coarse and fine registration. In the coarse step, a spanning tree can be constructed by applying the presented approach for pair-wise registration of partially overlapping scans with the genetic algorithm (GA). To construct the spanning tree, the dual criterion is proposed to judge the reliability of pair-wise registration results. In this case, the first scan can be selected as the root node and other scans can be added by the breadth-first search with the reliable results of the pair-wise registration. Subsequently, coarse results for multi-view registration can be calculated from the constructed spanning tree. In the fine step, the coarse registration results can be viewed as the initial parameters of the trimmed iterative closest point (TrICP) algorithm to obtain the accurate object model. Without any feature extraction, this approach can automatically achieve the multi-view registration of unordered range scans. Experiments were performed on public datasets to show its superiority on robustness for multi-view registration of unordered range scans. (C) 2015 Elsevier B.V. All rights reserved.