3D Point Cloud Initial Registration Using Surface Curvature and SURF Matching

Abstract

Initial registration of the 3D point clouds affects the final reconstruction result directly. However, some of the current point cloud initial registration methods are mainly based on the spatial distribution characteristics of the point clouds, which have the problems of large calculation, low accuracy of registration, and so on. In this paper, a new initial registration method based on the combination of texture features and curvature features is proposed to improve the speed and accuracy of the initial registration. First, the texture information of the point cloud model is projected onto a two-dimensional space to obtain its projection image. Then, the matched point pairs of the projection image are extracted using the SURF operator. Next, the curvature information of the matched point pairs is calculated, and only the point pairs with higher similarity are reserved. Finally, the transformed matrix is calculated using the reserved point pairs to achieve the initial registration. Experimental results show that the proposed method can implement the initial registration quickly and accurately.