Weld bead extraction based on point cloud segmentation

Abstract

At present, the welding seam extraction algorithm is for 2D space image processing, which has the disadvantages of complex extraction process, low accuracy and great limitations. Compared with the cumulative error caused by image mosaic, 3D can reflect the target features from various perspectives. At present, RANSAC fitting plane and Euclidean clustering extraction have been applied to 3D point cloud feature analysis, but RANSAC can only roughly fit the surface of the object, and it can not effectively extract features for the areas with rich surface curvature changes. Euclidean clustering is based on KD tree query where the algorithm needs to set the number of neighborhood points, so it has uncertainty. In order to overcome the above defects, this paper proposes a region growing segmentation algorithm, which can divide the region and extract the boundary according to the changes of the weld surface curvature and normal vector. At the same time, compared with the actual extraction results, the average error is reduced to about 1.0 mm. Compared with the traditional 2D laser line extraction, it can more intuitively reflect the weld features and put into practical production application.