Point cloud deep learning for multiple object pose estimation
Yukihiro TODA, Naoya Chiba, Koichi Hashimoto · The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) · 2019
6D pose estimation is an important task in the research field of robot vision. Pose estimation is used in picking task in industrial scene. the previous study shows that the pose estimation can be performed by deep learning if the scene is only a single object. In industrial applications, it is necessary to perform tasks in scenes multiple objects are placed in disorder. In multiple objects scene, each object must be identified from the scene, and pose estimation performed separately. In this paper we design a neural network that estimates multiple objects at the same time. Using an object clustering module, the network became possible to learn scenes of multiple objects.