Improved Tensor Voting And Its Application In Point Cloud Feature Estimation

With the development of three-dimensional measurement technology,point cloud processing has been more and more successfully applied in product reverse design,measurement-assisted digital medicine,autonomous navigation and other fields.Point cloud has gradually become a new data form after time series and images.As an important branch of point cloud data processing,feature estimation has an important impact on smoothing filtering,scene segmentation,parameter identification and other post-processing of point cloud data.Based on tensor voting theory,this paper explores a new mechanism of anisotropic analytic tensor voting,and proposes a fast and accurate estimation method of point cloud features based on anisotropic analytic tensor voting algorithm.The research contents are as follows:Firstly,aiming at the complexity and inefficiency of traditional tensor voting algorithm,a new analytical tensor voting mechanism is proposed.A new tensor voting attenuation function is designed to lay a foundation for the analytical solution of tensor voting.The new tensor voting attenuation function is applied to obtain the analytical solution of bar tensor voting.The evolution mechanism of bar tensor to plate tensor and ball tensor is explored.By constructing vector functions with controllable parameters,the analytical solutions of plate tensor and ball tensor voting are obtained,which lays a foundation for fast and robust estimation of point cloud characteristics.Basics.Secondly,a new anisotropic analytical tensor voting mechanism is proposed to solve the problem that the angle sensitivity of the attenuation function of analytical tensor voting is poor and the local structure of point cloud has insufficient influence on the voting process.The mapping relationship between tensor features and differential geometric features of point cloud surface is analyzed;a new anisotropic tensor voting attenuation function is constructed;based on the anisotropic tensor voting attenuation function,a new mechanism of anisotropic analytical tensor voting is constructed,and the analytical solutions of anisotropic analytical bar tensor,plate tensor and spherical tensor voting are obtained,which lays the foundation for fast,accurate and robust estimation of point cloudfeatures.Foundation.Then,aiming at the low efficiency of normal,curvature estimation and feature extraction of point cloud data,a comparative experiment of normal and curvature estimation of point cloud data based on anisotropic analytic tensor voting is constructed.It lays a foundation for fast and accurate estimation of normal and curvature features of point cloud data,and builds a feature point extraction model based on anisotropic analytic tensor voting,which lays a foundation for fast and reliable feature extraction of point cloud data.Finally,an experimental study on feature estimation of point cloud data is carried out to verify the performance of the proposed point cloud feature estimation method based on anisotropic analytic tensor voting in terms of accuracy,efficiency and robustness of feature estimation.The experimental comparison shows that the efficiency of the normal curvature estimation based on this algorithm is 10 times faster than that of the comparison method,and the anti-noise ability is similar to that of the traditional method.Feature extraction is more accurate and efficient than traditional methods.The more data,the more obvious.As far as robustness is concerned,within the noise ratio of 1:3,the extraction rate of feature points in this paper is more than 90%.