Recognizing Articulated Objects In Robotic Tasks

Abstract

The paper presents a graph-based method for articulated object recognition and handling using skeleton computation. Handling articulated objects is a difficult task because one cannot define a default grasping position relative to the centre of mass for a recognized object. The method presented in the following pages is a possible approach in automated assembly/sorting tasks using industrial vision-robot systems for articulated objects where the articulated objects can be recognized irrespective to the different possible angles between theirs articulated segments and can be grasped without any risk of collision.
PDF