Point Cloud Normal Estimation Via The Accumulation Of Difference Based On Geodesic Path

With the development of 3 d scanners,people can easily get all kinds of original point cloud data at the same time,the reverse engineering a variety of domains such as industrial manufacturing more and more extensive application of the original point cloud data important geometric properties of the normal vector is a point cloud,fast accurate method to estimate is the basis of the point cloud data analysis and processing of the reconstruction of the point cloud rendering of anisotropic smooth segmentation feature detection and extraction,and many other applications,rely on high quality method to the information,however,for the sharp feature points,the neighborhood is composed of joining together of two or more smooth surfaces,This makes accurate normal estimation very difficult.If the estimation is not accurate,the detailed features of the surface are easy to be lost in point cloud processing,and the reconstructed surface is difficult to recover the geometric features of the original model.Segmentation based method to estimate algorithm is mainly through the initial method to the point of difference to construct the similarity between.Because of the lacking of distance attribute,this kind of algorithm for estimating result is close to the surface and smooth surface is not ideal in this paper,a method to estimate algorithm based on the differences of the short circuit,using the most short-circuit the initial normal difference and the location of the point of information fusion,short-circuit point by superimposing the initial normal differences,calculating the similarity between the two so calculation of similarity to consider not only differences between normal,In addition,the structure information contained in the shortest circuit itself is effectively utilized,and the changes of point normal in the path are taken into account to effectively characterize the adjacent structures of the adjacent surface and the smooth surface.The specific steps of the algorithm in this paper are as follows: first,for the neighborhood of some points,find the shortest path between the neighborhood points,and calculate the similarity between the points by superposing the initial normal difference of the midpoint of the shortest path.Then,the neighborhood of candidate feature points is divided into several smooth sub-neighborhoods by spectral segmentation,and the optimal sub-neighborhood is selected to estimate the normal of candidate feature points.Finally,in order to improve the efficiency,a subspace structure propagation algorithm with normal constraints is proposed,which only needs to realize neighborhood segmentation of somepoints through the difference accumulation algorithm,and the neighborhood segmentation results of the remaining points can be inferred from the existing segmentation results.Experimental results on simulation and real scan data show that the proposed algorithm is capable of overcoming noise and anisotropic samplings,while preserving sharp features within the original point data.