A REAL-TIME EMG DRIVEN VIRTUAL PROSTHESIS HAND

This research work proposes a novel EMG pattern recognition to drive a virtual prosthesis hand from a two-channel surface EMG signal. A set of hand movements (four movements) is recorded with surface electrodes from the forearm of an able-body subject. Each movement is repeated and recorded ten times. Several computationally costly methods have been proposed and applied to EMG signal to generate classification information. In this paper, a computationally inexpensive method is devised to discriminate a reduced set of movements (in real-time). As a result of this approach, the following steps are implemented: i) high-frequency noise is removed from the EMG signals, ii) the EMG signals are rectified, iii) envelope of the signal is obtained by a low-pass filter, iv) the energy level of each envelope signal is computed and a different threshold level is set for each EMG-channel to decide whether the muscle is active or inactive, and, v) active muscle is labeled as logic high (1) and inactive muscle is presented as logic low (0). This logical presentation of muscle activities allows us to generate four distinct classes of hand movements and drive a virtual hand in real-time. Acceptingly, there is a trade-off between computational complexity of applied methodes) and discriminating ability of the algorithm. The ability of the proposed algorithm is employed to drive a virtual hand.