Recognition of Human Hand Motions for Robot Learning by Observation

The authors introduce you in this paper, not the method of imitation learning itself, but the method of hand pose estimation with single or dual high-speed camera. We believe that hand pose estimation at high speed and with high accuracy, i.e. recognition of human hand motions, is essential for robot learning by observation or its imitation learning. The purpose of this paper is to propose a remotecontrolled robot system capable of accurate and high-speed performance of the same operation in strict conformity to the movement of the human operator, without sensors being installed or special control means being used. In particular, this paper intends to introduce a method for implementing a high-precision 3D finger pose estimation at a high speed that permits real time operation of a remote-controlled robot by two cameras installed at positions of loosely orthogonal relationship, using one PC of the normal specifications.