Comparison of Initial Registration Algorithms Suitable for ICP Algorithm

Abstract

In order to find a point cloud initial registration algorithm more suitable for practical applications. This paper selects the representative SAC-IA algorithm, 4PCS algorithm and NDT algorithm from the initial registration algorithm based on local feature descriptor, the registration algorithm based on global search strategy and the registration algorithm based on probability distribution. This paper verifies the three initial registration algorithms, and records the registration time of the point cloud and the rotation error and translation error on the XYZ axis after the point cloud registration in the experimental verification process. By analyzing the experimental results and comparing the registration efficiency and registration accuracy of different point cloud registration algorithms, the most suitable initial registration algorithm is obtained. Finally, according to the initial registration results, the classic ICP algorithm is used to accurately register the point cloud. The experiment proves that the efficiency and accuracy of the SAC_IA algorithm based on the FPFH feature descriptor are generally better; the registration accuracy of the 4PCS algorithm is better, but the algorithm runs longer and the efficiency is not high; the experimental results of the NDT algorithm are more general, and the registration accuracy In terms of efficiency and efficiency, there is no advantage of the SAC_IA and 4PCS algorithms.