Study On 3D Laser Point Cloud Data Denoising Method Based On Clustering Algorithm

With the rapid development of 3D laser scanning technology,information science and computer science,it has become a research hotspot to obtain point cloud information of actual objects through 3D laser scanners and use them for model reconstruction.Its applications are very extensive,such as the protection and restoration of cultural relics and historic sites,the development of 3D game characters,the construction of models in virtual reality,and the design and restoration of various molds.In the process of establishing a 3D laser point cloud data model,point cloud data preprocessing is one of the most important links.Point cloud data preprocessing mainly includes denoising,streamlining,and vulnerability patching.This article mainly focuses on point cloud denoising.(1)Neighborhood search,normal and curvature estimation of scattered point cloud data.To pre-process the scattered 3D laser point cloud data obtained,it is necessary to establish a topological relationship for the scattered 3D laser point cloud data,perform a neighborhood search,and perform normal estimation.First,the KD tree method is used to establish the topological relationship of the scattered point cloud data,and then the geometric information of the point cloud data is calculated.Secondly,the normal vector of the point cloud data is estimated by the principal component analysis method.The direction of the normal vector is adjusted,and the curvature of the point cloud data is estimated by using the fitting method of the local quadric surface.(2)Research on clustering algorithms.For the noise points and outliers in the obtained point cloud data,two clustering algorithms,K-Means and DBSCAN,were selected to denoise the point cloud data.The experiment proves that the K-Means clustering algorithm is a distance-based clustering algorithm,and its algorithm is more efficient in processing data.The DBSCAN clustering algorithm is a density-based clustering algorithm that can cluster data of any shape.Both clustering algorithms have a good classification effect,which can effectively remove noise points and outliers.(3)Research on filtering algorithms.In the denoising process of scattered three-dimensional point cloud data,due to the huge amount of point cloud data,it is difficult to maintain the feature information of many point cloud data.Aiming at this phenomenon,this paper proposes a bilateral filtering method based on the normal vector modification.This method uses the adjusted normal vector to perform bilateral filtering,which can better maintain the boundary feature information.(4)Research on denoising methods of point cloud data.This paper aims at the problem of noise points in the scattered 3D laser point cloud data obtained by Shenyang Jianzhu University's old school gates,the Big Eight Kings Academy with large data amounts,and the Xinning Science Hall with large surface changes.Means clustering normal vector correction bilateral filtering fusion method and DBSCAN clustering normal vector correction bilateral filtering fusion method respectively cluster and smooth denoise solid point cloud data.Through different algorithms and experiments,the effects of different algorithms in different experiments are compared and analyzed.The results show that the K-Means clustering based normal vector correction bilateral filtering fusion method and the DBSCAN clustering and normal vector correction bilateral filtering fusion method Can effectively remove noise points.Among them,the normal vector modified bilateral filtering fusion method based on K-Means clustering is faster and more suitable for large entities,and the normal vector modified bilateral filtering fusion method based on DBSCAN clustering is more suitable for high-quality,small and complex real objects.