Hull shape laser point cloud registration method based on improved iterative nearest point algorithm

Abstract

Aiming at the point cloud data of hull shape obtained from laser scanner, which is characterized by large scale and many noise points, as well as the problems of low registration efficiency, slow convergence speed and poor robustness of traditional point cloud registration algorithms, a complex hull shape laser point cloud registration method based on improved iterative nearest point algorithm (ICP) is proposed. Firstly, a filtering method is used to denoise and the voxel filter is used for downsampling. After that, the key points are extracted from the hull shape point cloud by Intrinsic Shape Signatures algorithm (ISS), further construct the Fast Point Feature Histogram (FPFH) and find the corresponding point pair relationship, and use the random sampling consistency algorithm to filter the incorrect corresponding point pairs. Then, the Quaternion method is used to calculate the transformation parameters to complete the initial registration. The k-dimensional tree structure is established to enhance the operational efficiency of ICP algorithms, and aiming at the problem that ICP algorithm converges slowly and is prone to incorrect corresponding point pairs, the point-to-plane ICP algorithm by combining with distance constraints and normal vector included angle constraints proposed for fine alignment. The experiment indicated that in the registration experiments of ship hull laser point cloud data, compared with the traditional ICP algorithm, the proposed method has a higher registration efficiency and shortens the registration times, and can efficiently and accurately complete the large-scale ship hull shape point cloud registration.