GMM modeling of person information from EMG signals

Abstract

Preliminary results indicate that with well-chosen feature extraction, and modeling approach an identification rate of up to 97.9592% is achievable This paper proposes a vector quantization (VQ) and Gaussian mixture model (GMM) for modeling electromyogram (EMG) signal. A direct connection between muscle, central nervous system and brain is unique to an individual. The proposed method is applied to the electromyogram (EMG) pattern recognition for person identification, and experiments conducted to recognize forty nine subjects from EMG signals. The proposed learning method effectively identifies the change of feature vectors according to the subject and the GMM demonstrated high person identification accuracy. Our approach consists of a robust feature extraction scheme which is based on non uniform filter bank combined with VQ and GMM modeling. Various experiments have been conducted to determine the person identification performance in our proposed scheme for a database consisting of forty nine individuals, with collected 3 sessions EMG data in a gap of one day duration.