3D wrist gesture recognition using three-dimensional directional codes and HMM

Abstract

This paper proposes a novel 3D wrist gesture recognition method as the human-machine interface for mobile robot control. A sequence of depth maps was extracted from Kinect sensor on wrist gestures of an operator and a wrist gesture was specified by a sequence of 24-directional codes, which represents 3D motions with feature values of a single type, including directional and depth information. HMM algorithm was used to recognize wrist gestures, in which a fully connected hidden Markov model was designed and trained by Viterbi method and Baum-Welch method and the forward-backward procedure was used for the likelihood evaluation. The performance evaluation considering various factors showed above 96% of recognition rate and higher recognition rate on wrist gestures with complex motions.