主催: 一般社団法人 日本機械学会
会議名: ロボティクス・メカトロニクス 講演会2020
開催日: 2020/05/27 - 2020/05/30
In recent years, there are many multi-finger robots which can manipulate rigid objects, but it is difficult to manipulate flexible objects such as paper. Our research group developed a multi-finger robot system and succeeded in folding paper. In this system, the paper was recognized using RGB-D sensor and physical model. However, during occlusion, the matching between point cloud and physical model was unstable. In this paper, we propose a new method for recognizing a paper shape. First, a paper contour is detected using a RGB image. Next, point cloud data is obtained from a depth image. Finally, the paper shape is estimated by integrating the information from the contour, the point cloud, and a physical model. The experimental results show the effectiveness of the proposed method during occlusion.