[4Rin1] Interactive 2

[4Rin1-93] Automatic annotation of normal direction and invisible area of packaged objects for random picking

〇Takuya Kiyokawa¹, Jun Takamatsu¹, Tsukasa Ogasawara¹ (1.Nara Institute of Science and Technology)

Keywords:Picking robot, Object detection, Automatic annotation

To quickly deploy a vision system for random picking of packaged objects using a suction gripper, in this paper, we proposed an automatic dataset collection method for the vision system needed for random picking. As a result of experiments of random picking using our vision system, the detection accuracy shows that the F-measure value is 83% in the detection of five types of packaged snacks. The developed random picking robot system succeeded in picking the target objects with a small number of failures.
The results demonstrate the usefulness of constructing a detector for packaged objects with the proposed automatic dataset collection method.