An Improved Morphological Algorithm for Defect Detection on Point Cloud Data

There are often various defect points in the original point cloud data. Therefore, the point cloud data need to be pre-processed by filtering algorithms. The classical morphological algorithm is a common method to filter point cloud data. However, classical morphological algorithms often have limitations when applied to large-scale filtering. On the one hand, the classical algorithm adopts single opening or single closing operation to filter the point cloud data, which results in its inability to detect concave and convex defect points simultaneously. On the other hand, the filtering window linear increment strategy is used in the classical algorithm, which may influence filtering accuracy when applied to large workpieces. Because of these issues, this paper proposes two methods to improve the classical morphology algorithm and applies it to filter point cloud data on the rotor surface. One method is to change single opening or single closing operation to alternating opening and closing operations to realize simultaneous detection of concave and convex defects, while the other is to design a more flexible filter window increment strategy based on rotor surface curvature. Finally, compared with the classical algorithm, we show that the filtering accuracy can be increased by more than 11% through the improved algorithm.