Box-like Superquadric Recovery in Range Images by Fusing Region and Boundary Information

Abstract

This work contributes to the robotic bin-picking problem, and more specifically to the problem of localizing piled box-like objects. We employ range imagery, and use box-like superquadrics for modeling the target objects. Our approach for superquadric segmentation is an extension of the widespread recover-and-select framework, which employs only region information and therefore suffers from the region over-growing problem. Our approach equally considers both region and boundary-based information for performing the recovery task. Extensive experimentation with a variety of target object configurations demonstrates that it outperforms the recover-and-select framework in terms of both robustness and computational efficiency. Moreover, if implemented in a parallel hardware environment, our approach can operate in real time