Research on Point Cloud Registration Algorithm Based on Gaussian Curvature

摘要

Aiming at the problems of the Iterative Closest Point (ICP) registration with time consuming, poor robustness and low overlap rate point cloud registration difficulties, an improved ICP algorithm based Gauss Theorema Egregium that Gaussian curvature remains unchanged under the isocratic transformation of the surface in $R^{3}$ is proposed. Firstly, setting a Gaussian curvature threshold removes outliers and simplifies point cloud data. The simplified data is further divided into planar points and corner points. Next, the Random Sampling Consistency (RANSAC) algorithm is used for the corner points to achieve the purpose of rough registration. Finally, the K-dimension tree nearest neighbor search strategy is introduced to accelerate the search for corresponding point pairs, and the directional threshold method is used to remove the mistake point pairs to achieve accurate registration. Experimental results show that the algorithm in this paper not only has higher efficiency but also stronger robustness and more accurate than ICP algorithm.