Research On 3D Point Cloud Data Acquisition And Reconstruction Method For Line Structured Light Images Of Weld

In recent years,with the rapid development of robotics and the large-scale popularity of industrial robots,the use of robots to replace manual grinding of complex welds has inevitably become the future development trend.Grinding and polishing robots can significantly improve the accuracy,speed,and quality of grinding and guide the whole grinding process through force position control in grinding,reducing the damage to the base material,improving the work efficiency,and reducing the risk of accidental injury occurrence.However,the emergence of problems such as the location,height,width,and height of different welds and the presence of defects in the weld makes it impossible to have universal applicability by using a demonstration method to guide the robot for grinding and polishing operations.For this reason,it is necessary to use the vision measurement technology in computer vision to enable the grinding and polishing robot to obtain the weld profile characteristics in advance and then perform the grinding work for the specific weld.This paper is based on the active vision measurement technology of line structured light scanning.First,use a calibrated 3D data acquisition experimental platform to acquire many images of welds with line structured light stripes projected by direct laser triangulation onto the weld samples.Subsequently,use a fast extraction method of line structured light centerline to extract the centerline of images and obtain sub-pixel accuracy coordinates of structured light stripe centerline with weld morphology features in the 2D pixel coordinate system,which combines the weighted squared grayscale center of gravity method and singular value decomposition method based on the idea of data dimensionality reduction.Then,construct a line surface model by projection transformation to convert 2D pixel points to 3D spatial points in the camera coordinate system to obtain 3D point cloud data of the target weld sample.Then,dimensionally measure and compare the initial point cloud of weld features with the traditional manual measurement method using weld inspection tape to verify the improvement of measurement accuracy and speed.Then,use a point cloud de-noising method combined with statistical filtering,moving least squares method,and the voxel filtering down-sampling to filter and de-noise the different types of noisy point clouds generated within the point cloud.Remove a large amount of redundant data to improve the computational efficiency of the surface reconstruction process while retaining the weld morphological features to the maximum extent.Finally,use a greedy projection triangulation surface reconstruction method based on Delaunay triangulation to establish the triangular mesh topology by the Delaunay triangulation criterion.Subject the reconstructed surface to Laplace smoothing operation to complete the surface reconstruction and visualization for the weld samples.Compare the accuracy of the reconstructed 3D surface with the collected 3D point cloud data.The experimental results show that the experimental platform built in this paper meets the actual measurement accuracy,the proposed light stripe centerline extraction method can balance extraction speed and accuracy,the collected weld 3D point cloud data has less noise,and the final reconstructed triangular mesh surface can effectively reflect the actual 3D morphological characteristics of the weld sample,thus laying the foundation for future research related to force-controlled weld grinding in combination with robotics,laying the foundation for future research related to force-controlled weld grinding in combination with robotics.