Dynamic Hand Gesture Recognition using Sequence of Human Joint Relative Angles

Abstract

Hand gestures have been used as a natural and spontaneous form of nonverbal communication since the beginning of humanity. The interest in this field of study is expanding as a result of recent research endeavors. The method for dynamic hand gesture identification in this paper is based on a 3D skeletal model and uses depth pictures. The series of spatiotemporal changes in the relative angles of several skeletal joints with respect to a reference joint is used to suggest a new gesture representation. Over a predetermined number of frames, a series of significant Joint Relative Angles (JRA) between two skeletal joints is calculated. We identified a collection of 12 dynamic gestures with 98.6% accuracy using machine learning algorithms to analyze this sequential data.