Research On Target Edge Detection Based On 3D Point Cloud

In recent years,the rapid development of laser scanning systems and sensor technology has made 3D point cloud data acquisition more and more convenient,and the application of point cloud processing technology has become more and more extensive.Extracting effective edge description features from massive point clouds is to further improve machine vision applications Accuracy creates the possibility,so point cloud edge detection has increasingly become an important research direction in point cloud processing technology.Therefore,this article takes the edge detection of the point cloud model as the key research content,and the main work of this paper is as follows:(1)Aiming at the problems of the existing point cloud feature descriptors with high dimensions,complex calculations,and large amount of point cloud,a local edge feature descriptor(LEFD)based on curvature density is proposed,which strengthens the local expression of point clouds.This descriptor can effectively distinguish feature points from non-feature points.Substituting the average density of the neighborhood for the point density enhances the robustness of the descriptor.Aiming at the problem that the setting of the descriptor threshold is difficult to control,this paper automatically finds the optimal solution according to the LEFD feature distribution law of the input point cloud,which reduces the error caused by manual tuning.The experiment proves that the descriptor has certain robustness and strong edge description ability.(2)Aiming at the problem that the existing edge detection algorithm sets fixed neighborhoods when calculating point features,resulting in inaccurate calculation of point features,an automatic neighborhood selection method based on normal vector change rate is proposed.The algorithm can automatically set the best neighborhood according to the MANV change rate of the local position of the query point,and improve the accuracy of geometric feature calculation.Aiming at the problem that the existing edge detection algorithm is not highly automated,this paper proposes an edge detection algorithm based on the feature descriptor combining the neighborhood adaptive selection algorithm and the LEFD descriptor.The distribution characteristics of the adaptive adjustment of the relevant parameters.Experiments prove that the algorithm has certain adaptability to the input point cloud model.(3)Aiming at the wheel-rail contact point cloud data lacking a standard model,we constructed a wheel and rail point cloud data set.In the preprocessing stage of the wheelrail point cloud,a non-uniform sampling method based on voxel grids and curvature features is proposed to solve the problem that the classic point cloud reduction algorithm cannot retain the sharp features of the point cloud.By dividing the grid,the simplified point cloud can avoid voids in the flat area,and at the same time ensure that the data points of the feature area are retained to a large extent.The experiment proves that the difference between the point cloud feature area and the non-feature area after using this method is more obvious,which is good for the edge feature extraction work.It also proves that the edge detection algorithm in this paper has certain reference value and significance for the measurement of the wheel-rail contact relationship.