An Improved Iterative Closest Point (ICP) Method with Subsets
Xuefei Li1, Huimin Feng2, Qiang He3*
1College of Science, Agricultural University of Hebei, Baoding, 071001, China
2Key Lab of Machine Learning and Computational Intelligence, College of Mathematics and
Information Sciences, Hebei University, Baoding, 071002, China
3School of Science, Beijing University of Civil Engineering and Architecture, Beijing, 100044, China
Abstract: Iterative Closest Point (ICP) is a conventional, rigid point-sets registration approach with squared order complexity. It and its numerous variants have been widely applied in computer vision, and image processing fields. However, registration performance is degraded as cardinal of the point-set increases, which makes them impractical for large point-sets. In this paper, a subset based Iterative Closest Point method, namely subset-ICP, is proposed to reduce computational cost and registration error for large point-sets. Streaming method manages to obtain subsets of two point-sets. Registration then is performed on subset pairs from the two sets to determine Euclidean transformation parameters before they are used to move one point-set to approach to the other point-set. The usage of subset benefits subset-ICP by efficiently dealing with large point-sets registration, simultaneously capturing local deformation. Six public, available point datasets are used in experiment to examine the proposed method, to compare with the other three ICP variants and rigid CPD method. Experimental results show that the proposed subset-ICP is an efficient rigid point-sets registration method at sufficiently comparable level with the original method, especially on large point-sets.
Keywords: Rigid registration; Point-set; Iterative closest point (ICP); Transformation; Principal component analysis (PCA)