Point cloud registration algorithm based on normal vector and gaussian curvature

Iterative closest point(ICP) algorithm is widely used due to its high registration accuracy and strong adaptability. However, it is easily affected byGaussian noise and outliers, resulting in slow running speedandreduced registration accuracy, whichrequires that the point cloud must havea good initial position, otherwise local optimization problems will occur. In view ofthe above problems, this paper proposes a new point cloud registration method, which uses coarse vector and Gaussian curvature for coarse registration, removes irrelevant points and provides a good initial position, and then uses singular value decomposition based ICP algorithm to refine registration, and use Stanford′s point cloud data to set for registration experiments. The results show that the algorithm in this paper can effectively reduce the interference of Gaussian noise and outliers on the registration effect and improve the operation efficiency and registration of point cloud registration. Compared with the traditional ICP algorithm, the average registration time is reduced by 53.5% and the registration accuracy is increased by 43.2%.