Segmentation and approximation of objects in pointclouds using superquadrics

Abstract

The abstraction from raw pointclouds to high level representations is a necessity in real world robotic applications. This concerns the semantic description of household objects as well as the handling of complex objects (like grasping). This paper describes a method for the manual segmentation of objects from scenes and the automated approximation of objects with high level descriptors. For both tasks solutions based on superquadrics are presented.