Forearm Muscle Synergy Reducing Dimension of the Feature Matrix in Hand Gesture Recognition

Abstract

Robotic hand exoskeleton has been proposed to train the paretic hand after stroke, and its movements can be controlled by recognized activities of non-paretic hand. Although a variety of myoelectric pattern identification strategies have been proposed to distinguish different hand gestures by the surface electromyogram (sEMG) signals, the feature matrix of hand movements should be simplified for on-line myoelectric control. Synergistic muscle activity has been known as a principle for center nervous system (CNS) to modulate the multi-joint limb movement, here, we aim to study the feasibility of the muscle synergies for gesture recognition with simplified feature vector. Five healthy participants were recruited to perform hand tasks of hand open(open), hand close(close), key pinch(key), palm valgus(val), and grasp cylindrical tool(cyl). sEMG were recorded from six forearm hand muscles, and non-negative matrix factorization (NMF) was used to extract the synergistic patterns of myoelectrical activities. The preliminary results showed that each hand gesture can be expressed with one muscle synergy pattern, which consists of the contribution of six forearm muscles. With the synergistic features, the recognition rate of hand gestures reached 96.08%. The present work suggested that the muscle synergy can be used for hand gesture recognition with low dimension feature vector.