Research On Point Cloud Data Reduction Method For Preserving Features

With the rapid development of laser scanning technology and computer technology,threedimensional laser scanner can obtain a large number of high-precision point cloud models in a short time.However,these high-precision point cloud models contain a large number of redundant data,which greatly affects the three-dimensional surface modeling and feature judgment.Feature points are the basic feature primitives of surface geometry,and are also the key data points of simplified data in model reconstruction.The quality of feature extraction directly affects the shape expression and matching accuracy of subsequent models.In this paper,the point cloud model is taken as the research object,and the feature extraction method based on normal vector for scattered point cloud data is studied.The feature preservation method is applied to data reduction in order to improve the quality of model reconstruction after simplification.The main research contents and conclusions are as follows:(1)The neighborhood of three-dimensional point cloud data mainly includes K neighborhood,Voronoi neighborhood and BSP neighborhood.K-neighborhood calculation needs to be judged by point cloud model,point cloud density and uniformity,so this paper chooses more reliable K-neighborhood in data processing.In order to facilitate the construction of neighborhood,it is necessary to establish the point cloud data topological relationship first.This paper analyses three common methods to establish the point cloud data topological relationship,and studies the specific steps of the octree method with higher efficiency in the construction of k-neighborhood.(2)The traditional method based on local surface fitting is sensitive to the size of neighborhood,and the estimation of normal vector is inaccurate,which leads to misjudgement when extracting feature points from normal vector information.In order to solve this problem,a normal vector estimation method based on moving least squares fitting local surface is studied in this paper.Firstly,the local surface is fitted by moving least squares on the basis of constructing k-neighborhood,and then the normal vector of point cloud is estimated based on the fitted local surface,and the normal vector direction is unified.Finally,the average cosine of the angle between the sampling point and K-neighborhood normal vector is taken as the eigenvalue,and the feature points in point cloud data are extracted by comparing the eigenvalue with the threshold value.Experiments show that within the appropriate range of neighborhood selection of the model,the method in this paper is less affected by neighborhood points when solving normal vectors.Compared with traditional methods,it has higher accuracy and stronger robustness.(3)The existing simplification methods are not easy to preserve the original data features when dealing with point cloud models with large surface changes,which results in a large deviation between the reconstructed model and the original model.In order to solve this problem,this paper proposes a data reduction method based on feature preservation by analyzing and researching several common methods.Firstly,the feature points in the original data are determined and kept unchanged as the core points.Then,the bounding box is built in the non-feature area by using the center of gravity compression algorithm of the bounding box to simplify the corresponding proportion of data.Finally,the extracted feature points and the reduced non-feature points set by the center of gravity compression method of bounding box are output.Experiments show that the simplified point cloud data retains the original feature points while realizing the simplification.Comparing different simplification methods to reconstruct model and standard deviation of model,it shows that this method has certain advantages over traditional methods in data reduction.