PUB - Publikationen an der Universität Bielefeld

Industrial part assembly has come a long way and so has visual quality inspection. Nevertheless, the key issue in automated industrial quality inspection, i.e. the pose recovery of the objects under inspection, is still a challenging task for assemblies with more than two rigid parts. This paper presents a system for the pose recovery of assemblies consisting of an arbitrary number of rigid subparts. In an offline stage, the system extracts edge information from CAD models. Online, the system uses a novel kernel particle filter to recover the full pose of the visible subparts of the assembly under inspection. The accuracy of the pose estimation is evaluated and compared to state-of-the-art systems.