"Matching-Tree - A Novel Data Structure for Global Automatic Matching

Matching of multiple views is normally consists of coarse matching, fine matching, and a bundle adjustment at the end. Coarse matching, that is the pre-alignment of the surfaces of the complex forms, which can be positioned far away from each other in 3D space, is a difficult problem to solve. Fine matching on the other hand can be performed accurately using either the ICP (iterative closest point) method or the least square surface matching method. At the end, a bundle adjustment aligns the multiple views exactly. To perform the automatic coarse matching, this paper presents a fast algorithm to construct the “matching tree”, which was innovated by the authors in the earlier paper [1]. The complexity of the algorithms is dropped from ( ) 4 n Ο to ( ) 2 n Ο , and the real run time is accelerated by factor 100. Validation of the method has been performed based on reconstructing the Castle Neuschwanstein out of laser range scanner data.