Reliable recovery of piled box-like objects via parabolically deformable superquadrics

Automatic unloading of piled box-like objects is undoubtedly of great importance to the industry. In this contribution a system addressing this problem is described: we employ a laser range finder for data acquisition, and globally deformable superquadrics for object modeling. Our technique is based on a hypothesis generation and refinement scheme. The vertices of the piled objects are extracted and superquadric seeds are aligned at these vertices. The model parameter recovery task is decomposed into two subproblems, each dealing with a subset of the model´s parameter set. Both region and boundary based information sources are used for parameter estimation. Compared to a widespread strategy for superquadric recovery, our method shows advantages in terms of robustness and computational efficiency. In addition, our system exhibits versatility with regard to existing industrial systems, since it can effectively deal with both neatly placed and jumbled configurations of objects.