Gesture Recognition via Estimation of Information Exchange between Muscles

A new approach based on exploitation of the information exchanged by human muscles is proposed for recognition of the resulting hand gesture. Such information can be realized through estimation of the causality between the multichannel surface electromyography (sEMG) signals over short signal windows. These measures can then be classified effectively not only to accurately identify the hand gesture but also to study the human motion dynamics, which can be effectively used for rehabilitation purposes. In our experiment, we see that this approach can achieve an overall classification accuracy as high as 90.3% even with a small data size, which can be hardly achieved by traditional methods with the same classifiers.