The Iterative Closest Point Registration Algorithm Based on the Normal Distribution Transformation

Abstract

Traditional Iterative Closest Point (ICP) algorithm registration is slow, especially when the scale of the point cloud is relatively large. This paper proposes a new algorithm which is the iterative closest registration based on the normal distribution transform (NDT-ICP). The algorithm uses NDT as the coarse registration algorithm, because the NDT algorithm does not use the features of the corresponding points to calculate and match during the registration process, and does not need to consume a large price to find the nearest neighbor matching points, which speeds up the registration speed. The ICP algorithm is used as fine registration to further improve the accuracy of the overall registration. The results show that the NDT-ICP algorithm has obvious speed advantage while ensuring considerable accuracy. Especially for large-scale point cloud registration with few features, the accuracy of this algorithm is about twice that of SAC-IA-NDT algorithm, and slightly improved compared to SAC-IA-ICP algorithm. In terms of registration speed, the registration time of the NDT-ICP algorithm is about 10% and 11% of the SAC-IA-NDT algorithm and the SAC-IA-ICP algorithm, respectively. The experimental results show that the algorithm is efficient and effective.