Abstract: A self-adaptive iterative query algorithm based on mean reverse shift is proposed, which can be used to query topological neighbors for 3d scattered point-cloud. R*-tree is applied to organize the dynamic spatial index structure of scattered cloud-point. The topological neighboring reference points of object point are obtained by the k-nearest neighbor query algorithm based on R*-tree. According to the relationship between neighbors query and density distribution of the point set, kernel density estimation is used to describe the distribution law of point set. The mean shift vector and mean point which can reflect the local distribution features of sampling points are computed by mean shift algorithm. The topological neighbors self-adaptive iterative query algorithm is realized through the iterative search process by shifting neighbors search area along the reverse mean shift vector and querying new neighbors. The experimental results show that this method can obtain topological neighbors of arbitrary complicated scattered point-cloud efficiently, and the neighbors include Voronoi neighbors as well as some more available neighborhood reference data, which can reflect the local surface features around the object point better.

Key words: kernel density estimation, mean reverse shift, R*-tree, scattered point-cloud, topological neighbors query