キーポイント検出を用いた箱の把持位置検出

抄録

In production processes, there are works to collect parts from parts warehouses and parts shelves. Automation of this work is required. When the necessary parts are stored in parts boxes, the boxes must be detected and taken out from the shelves. In this paper, we use keypoint detection, a method based on deep learning, to detect grasping positions of boxes using an RGB-D camera. We set keypoints on the boxes for the learning. The grasping positions of boxes are detected by the keypoint detection model. The accuracy of this method is verified by grasping experiments.