An Automatic Registration Method of Scanned Point Clouds

To register scanned point clouds without any additional information,an automatic registration method is developed by introducing new metrics for matching corresponding points and modifying the iterative closest point algorithm.The whole registration consists of two steps: initial registration and fine registration.In the first step,based on the curvature similarity between two points,a new metric for the curvature similarity between the neighborhoods of them is developed to construct an initial array of corresponding points.Then two restrictions based on invariant features of the rigid transformation are introduced to pick out some corresponding points of high precision from the initial array.Initial registration parameters are computed directly according to the geometric information of these corresponding points.In the second step,the computing of the closest point is modified.Numerical experiments demonstrate the good results of the initial registration and the better results of the fine registration,which have met the requirement of registering point clouds from different views.