Recognition of 5-finger motion using EMG signals based on general object recognition
Daisuke Sakaguchi, Yunan He, Osamu Fukuda, Nobuhiko Yamaguchi, Hiroshi Okumura · The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) · 2019
In this paper, we propose a system to estimate 5-finger motions based on recognition of a general object and operator’s EMG signals. DCNN recognizes the object to be grasped by the operator using a camera image, and then LSTM network estimates the 5 finger joint angles from the EMG signals. LSTM networks are prepared for each target object and they are switched to use according to the recognition result of DCNN network. In order to verify estimation accuracy of the proposed system, we conducted the experiments on five types of target objects. As a result, we revealed that the 5 finger joint angles were successfully estimated with high accuracy. The estimation accuracy improved compared with the case where the system did not use the result of object recognition.