A real-time dynamic gesture recognition based on 3D trajectories in distinguishing similar gestures

There are many shape-similar gestures which cause errors in the process of hand gesture recognition. In this paper, a new method which can distinguish the similar gestures was proposed. The information of motion trajectory is captured by a leap motion in three-dimension space, and the orientation characteristics are quantified and coded as the feature. Then the Hidden Markov Model (HMM) algorithm is utilized to model and classify gestures. But there are many shape-similar gestures in our database (numbers 0-9 and alphabets A-Z) such as S and 5, Z and 2 whose recognition rates are low. In this paper, we proposed a new method that can distinguish the similar gestures in real-time. The experiment result demonstrates the effectiveness of the proposed method.