Edge detection from point cloud of worn parts

Abstract

3D scanners are able to quickly and accurately digitise objects into Point Cloud Data (PCD). It has been used in various applications, including damage identification for automated repair via additive manufacturing. Useful information, such as the geometrical edge information, has to be extracted from the PCD for damage identification. A common edge detection method is by thresholding high curvature points from a point cloud. However, edges on worn parts tend to have less distinct edges from wear. This would cause errors in curvature based edge detection such that a band of points is detected along the edge, instead of a single row of points. Other edge detection methods are also unable to accurately or robustly detect the worn edges. Hence, this paper seeks to solve the limitation of the state of the art of PCD based edge detection for detecting worn edges. In this paper, we present a method of detecting geometrical edges, which involves curvature thresholding, iterative non-maximal suppression, and feature line generation. The proposed method has been validated on a physically scanned part, and the results are presented.