Recognizing Hand Gestures of a Dancer

Abstract

A new and simple two-level decision making system has been designed for performing scale-, translation- and rotation-invariant recognition of various single-hand gestures of a dancer. The orientation filter is used at the first-level to generate a feature vector that is able to distinguish between several gestures. At the second-level the silhouette of the different gestures is extracted, followed by the generation of the corresponding skeleton and the evaluation of the gradients at its end points. These gradients constitute the second feature set, for recognizing those gestures which remain to be identified at the first-level. An application has been provided in the domain of single-hand gestures of Bharatanatyam, an Indian classical dance form.