| With the development of 3D scanning technology,it has been widely used in many research fields.Although the surface information of the measured object can be quickly obtained through three-dimensional laser technology,the original data obtained are three-dimensional space coordinates,which is scattered and contain a lot of noise,and must be denoised before application;Because of the overlarge amount of the original point cloud data obtained,and it requires higher hardware configuration and reduce the efficiency greatly when the original point cloud data are processed directly,so it is necessary to simplify the scattered point cloud in space.In addition,there are a large number of regular surface models in the real 3D scene,so in the later time of 3D reconstruction,the point cloud data which contained a large number of regular surface features should through the way of fitting processing first to obtain accurate model parameters.To sum up,the algorithms related to scattered point cloud data processing are researched deeply in this paper from three aspects:point cloud denoising,point cloud simplification and regular surface fitting.The main research work is as follows:1.The structure and geometric characteristics of scattered point cloud data are studied,and the search of point cloud neighborhood,curvature of point cloud and estimation of normal vector of point cloud are discussed respectively.On this basis,according to the characteristics of point cloud noise,the point cloud noise is divided into internal noise and external noise,and the large-scale external point cloud noise is eliminated by combining K-D trees algorithm and Gaussian filter.For the noise mixed in the main data,the random sampling consistent algorithm(RANSAC)is used to remove the noise due to the short distance between other target point clouds.Experiments show that this algorithm can not only eliminate the external noise point cloud data floating around the point cloud,but also effectively remove the point cloud noise in the main body,making the surface of the point cloud model smoother.2.In order to reduce the number of point clouds and the lack of target surface feature information in the process of point cloud simplification,this paper proposes a simplification method of multi-parameter evaluation of point cloud feature area,which combines the calculation of point cloud curvature and normal vector to check and retain the point cloud feature area through various evaluation parameters.Experiments show that this method can retain the feature details of the model to the maximum extent on the premise that the point cloud data is not empty.Experiments show that the point cloud simplification rates of Bunny and Dragon are different under different simplification conditions.In addition,the results of simplification are judged by the standard deviation and surface area change rate after model reconstruction in this paper.3.In order to overcome the problem of accuracy of fitting parameters in traditional fitting algorithms,this paper proposes to apply partial total least squares to point cloud fitting of regular surfaces,and use the parameter values estimated by RANSAC algorithm as the initial values of model iteration,and use adaptive modified weight matrix to optimize parameter estimation.The feasibility of this method is proved by simulation experiment.In order to further verify the reliability of this algorithm,globes with different radii are selected as experimental objects.The final experimental data show that the unit weight median error and fitting accuracy calculated by this algorithm are higher than those calculated by RWTLS algorithm.Figure[38]Table[16]Reference[90]... |
PDF Full Text Request