Research On Simplification Of Point Cloud With Preserved Features

In the fields of computer graphics,computer vision,image processing and reverse engineering,3D reconstruction technology has always been an important topic of research.With the continuous development of science and technology,three-dimensional laser scanning equipments such as the German Z+F three-dimensional laser scanner are constantly being updated,and the geometric information of the three-dimensional object can be saved to the storage device through the scanning device.The problem that comes with it is that there is a large amount of redundant data in the dense point cloud data.These redundant data greatly increase the storage overhead and computational cost of the point cloud data,and affect the surface fitting and model.Generate efficiency such as important follow-up work.It is also necessary to consider that point cloud modeling with different precisions requires original point cloud data with different precisions.Therefore,on the basis that the three-dimensional reconstruction model can retain the shape of the original three-dimensional object,the simplification of point cloud data becomes inevitable.The simplification of point cloud is a hot topic in the current 3D modeling research,which can facilitate subsequent work and improve the efficiency of 3D reconstruction.In this paper,the main research contents of the point cloud simplification are roughly divided into three main aspects: the calculation of point cloud data curvature,the point cloud data clustering method and the point cloud data simplification algorithm design:Firstly,in the calculation process of point cloud data curvature,since the original K-D tree neighbor search method takes a long time,each search backtracking algorithm is more complicated.This paper uses an improved K-D tree method to construct point cloud data.According to the topological relationship,the k neighbors of the point to be measured are quickly searched,and the parameter equations of the local surface are obtained by using the least squares method according to the k data points and the points to be measured,thereby calculating the curvature of the point cloud data.It is proved that the improved K-D tree method can quickly find k nearest neighbors and calculate the curvature value of the data points to be tested.Secondly,the classic K-Means clustering algorithm in the process of classifying point cloud data,the initial clustering center and the number of clusters will affect the clustering results.Aiming at this problem,this paper first establishes a K-D tree for the original point cloud data before clustering.According to the uniform distribution of data on both sides of the K-D tree subtree,the node with the number of nodes in the K-D tree closest to the user-specified k value is taken.For the cluster center,the number of clusters and the cluster center are determined accordingly.Experiments show that this method can not only solve the selection of clustering center,but also has faster convergence speed than the classic K-Means clustering algorithm in terms of running time.The simplification of point cloud data is the main research content of this paper.Combining the advantages of clustering reduction method and curvature reduction method,this paper firstly uses the improved K-Means clustering algorithm to cluster the point cloud data.Then,the curvature value of the point cloud data is sequentially calculated by using the K-D tree,and the feature area and the non-characteristic area are distinguished according to the curvature difference.In the feature region with complex surface changes,the feature points are identified by curvature information entropy and retained.The remaining points in the feature region cluster are reduced by random reduction;the non-feature regions retain the cluster center data points.Experiments show that this hybrid reduction method can preserve the detailed features of the point cloud raw data well,and there is no void in the non-feature area.