A novel torchlight data association strategy for surface registration

This paper presents a novel method for rigid surface registration using torchlight structure as data association, and the new method improves the correctness of point matching. When two sets of point clouds are merged, assume a set of torchlight beams parallely pass through them, and each light ray passes the overlapped data twice, one on each set. The Euclidean distance on such pair is taken as measurement of the separation. When the two sets are optimally aligned, the registration error is minimized. Hence, surface registration problem is reduced to a six degree of freedom searching procedure. Preprocessing, optimization, and acceleration modules are introduced to normalize raw data, explore registration space, and reduce execution time. Unlike the Iterative Closest Point (ICP) algorithm, the proposed approach does not require pre-alignment information. Secondly, the performance of ICP is poor when the overlapped area between two sets is not sufficiently large. The proposed approach does not suffer from these problems. Based on various experiments, the proposed approach shows the superior performance over ICP.