세 가지 다중 점 군 정합 방법 비교 분석에 대한 연구

Abstract

In this paper, three multi-view registration methods for point clouds, MAICP (Motion Averaging Iterative Closest Points), LUM with 6 DoF (Lu and Milios style SLAM with 6 DoF), and IOM (Improved Optimization-on-a-Manifold), are compared. For each method, three types of correspondence computation approaches: point-point, point-plane and point-curved surface, are implemented and the performance of the multi-view registration methods based on each correspondence computation type is analyzed and compared with real examples. To reduce the accumulative error caused by pairwise registration, multi-view registration should be applied. We also suggest which type of correspondence method, and the multi-view registration method should be applied to obtain the least error and registration time.