3D dynamic gesture recognition based on improved HMMs with entropy

Nowadays gesture recognition is a hot topic in the field of human-computer interaction (HCI). HCI develop very fast, and also brings surprise to us constantly. In this paper, we propose a novel approach based on improved HMMs with entropy to recognize the 3D gesture. In our method, there are two steps to recognize a gesture: 1. detect the key nodes of body with extracting the skeleton point. A low-pass filter is utilized to smooth trajectory later. 2. We use improved Hidden Markov Models (HMMs) algorithm which has a virtual start node and a virtual end node with another layer for gesture recognition. In order to decide when to start meaning gesture and when to end non-meaning gesture, we use entropy which can enlarge the searching space to avoid over-fitting and local minimum. Experimental results will demonstrate the performance of proposed approach.