Geometric Feature Detection for Reverse Engineering Using Range Imaging

Abstract The use of range imaging has been gaining popularity in reverse engineering. One challenging task is the detection of feature information from range images. In this paper, an approach to detect geometric features from range images using a fuzzy partitioning theory and geometric invariants is developed. Based on the fuzzy C-shell clustering technique, quadric features are partitioned into primitive clusters. Instead of performing sequential model fittings, general quadric surfaces as object shells are fitted concurrently. The geometric representations of prototypes are generated during the above process of pattern classifications. The integration of the partition with the invariant analysis makes it possible to detect geometric features from depth maps for the development of reverse engineering.