Computing superquadrics using binocular stereo sensor

The authors present an integrated approach to recovering a superquadrical model of a single object from stereo images. The algorithm combines the recovery processes of occluding contour, surface, and volumetric models in a cooperative and synergetic way. The aggregation of shape information is performed via model recovery. By integrating the depth information and orientation information from the respective surface and contour modules, the volumetric model based on superquadrics can be recovered robustly. This results in consistent interpretation of the object and an increase in system flexibility