Application of neighborhood feature in point cloud registration

Abstract

A new registration method of large-scale scattered point clouds based on invariant features of neighborhood was proposed,which consisted of preliminary registration and exact registration.Firstly,the target point set was weighted to reduce the amount of corresponding point-pairs efficiently.Secondly,on the basis of distance features between points and their neighborhood centroids,this paper added an additional geometric feature vector of included angle to eliminate bad point-pairs,and then the preliminary registration was completed.Finally,the Iterative Closest Point(ICP) algorithm with improved invariant feature was used to register accurately.The experimental results indicate the good results of the preliminary registration and the better results of the exact registration,which have met the requirement of registering point clouds from different viewpoints.