ソフトクラスタリングを用いたばら積み唐揚げの検出方法に関する検討

抄録

Since foods have irregular shapes, their individual recognition is difficult. We have previously proposed a food detection method for piled pieces of fried chicken. However, the method fails to detect a piece under certain circumstances due to a clustering error when the method handles a 3-dimentional point cloud. For the clustering process, we newly propose to use a soft clustering method in which each point can belong to more than one clusters if needed. The ambiguity does not become a problem for manipulation tasks and it will enhances the robustness of the clustering process. In this paper, we measure the improvement of accuracy with the new method.