Hole detection for industrial parts using inverse problem of persistent homology

Abstract

Persistent homology, which captures topological features of point clouds and has been used successfully in the field of topology in mathematics. This approach can extract shape features robustly by low-dimensional (i.e., 3D in this paper context) points. We applied this technique to 3D point cloud processing in robot vision. Persistent homology captures holes in point cloud as pairs of birth time and death time of changing radius of point distance. By selecting these pairs by criteria based on the focusing hole parameter, corresponding hole can be detected. In this paper, we demonstrate that a method called inverse analysis of persistent homology can be used to robustly detect holes in industrial parts.