Muscle-gesture robot hand control based on sEMG signals with wavelet transform features and neural network classifier

Abstract

In this paper, we propose a muscle gesture-computer interface (MGCI) system for a five-fingered robotic hand control employing a commercial wearable MYO gesture armband. Eight channels of surface EMG (sEMG) signals were acquired and segmented. Then four levels of Daubechies 5 Wavelet family were performed to analyze the EMG signal. Totally 72 features were extracted from the EMG raw data for 16 hand motions recognition utilizing artificial Neural Networks. The average of best overall classification rate during off-line training is 87.8%. Consequently, real-time hand gesture recognition was implemented to evaluate the performance of the proposed system and the average recognition accuracy was 89.38%. Finally, it was applied to control a five-fingered robot hand.