Feature extraction of laser point cloud and surface reconstruction for groove welds based on region adaptation

To address the problems of significant noise interference, low positioning accuracy of stripe centerlines, and insufficient surface reconstruction efficiency in the three-dimensional feature extraction of groove welds under complex lighting interference, a method for weld feature extraction based on laser point clouds was proposed. By constructing a point cloud acquisition platform, a compound filtering and denoising approach was combined with a gray centroid extraction method based on region automatic continuous centerline tracking to obtain laser stripe centerlines, which resolved the issue that the traditional gray centroid method failed to accurately locate stripes under uneven illumination conditions. Subsequently, region-adaptive sampling and feature enhancement were performed on the point cloud data derived from the extracted centerlines. Surface reconstruction was achieved by performing partitioned plane fitting for each region and applying Delaunay triangulation on the partitioned planes, thereby reconstructing the physical model of groove welds. Compared to the traditional Delaunay triangulation method, the surface reconstruction speed was improved by 73%. The results indicate that the maximum linear fitting error of weld feature points extracted by this method is less than 0.52 mm; the average relative error is below 1.54%; the MSE is less than 0.026 mm², and the R² exceeds 99.99%. This method provides reliable technical support for automated welding quality inspection and process optimization.