Research On Defect Detection Based On Edge Feature Point Cloud Registration

With the rapid development of computer vision and automation technology,the visual inspection technology on the surface defects of industrial products has become one of the hot research contents.The traditional two-dimensional image visual inspection technology realizes the defect detection on the surface of the object through the image texture,gray level and other information,but due to the lack of depth information in the two-dimensional image,this detection technology has limitations that cannot be ignored.Therefore,it is necessary and important to study the surface defect detection of objects using three-dimensional point cloud data.The defect is calculated by registering the standard point cloud and the point cloud to be inspected,and comparing the difference between the point clouds.The registration process will be optimized by using the edge features of the point cloud.The specific research content is as follows:In order to optimize the point cloud data and remove the noise in the point cloud data,the point cloud data preprocessing technology is researched.The voxel downsampling of the point cloud is performed first,which greatly reduces the amount of point cloud data while preserving the geometric characteristics of the point cloud.Then use the RANSAC(Random Sample Consensus)algorithm to eliminate the useless data,and only keep the key part of the point cloud of the object to be detected.Finally,according to the type of noise in the point cloud,statistical filtering and through filtering are used to filter out the outliers and the whole piece of noise in the point cloud data.By analyzing the performance defects of traditional point cloud coarse registration algorithm and point cloud fine registration algorithm,a point cloud registration method based on edge point cloud features is proposed.The method can be divided into several steps: calculate the edge point cloud of the source point cloud data and the standard point cloud data respectively,calculate the FPFH(Fast Point Feature Histogram)feature descriptor for the two pieces of edge point cloud data,and build according to the feature descriptor The corresponding point pairs of the two point clouds are used to complete the coarse registration calculation.Finally,the rough registration result is used as the initial point cloud of the Iterative Closest Point(ICP)algorithm,and the best rigid body transformation matrix is calculated by the ICP algorithm to complete the registration.Through experimental analysis,the speed and accuracy of point cloud registration have been greatly improved.Judge the defect points in the point cloud data after registration by searching the K neighborhood of the point cloud.This method first establishes the K neighborhood of the point in the standard point cloud to be detected,and calculates the Euclidean distance between the point and the nearest point in the K neighborhood to determine the defect point.In order to better summarize and display the defect point cloud,the greedy projection triangle mesh algorithm is used to complete the surface reconstruction of the defect point cloud,and the type of the defect point cloud is automatically classified.Experiments have proved that by searching the K neighborhood of the point cloud,the defect points in the point cloud to be inspected can be quickly found.