A novel method for simultaneous gesture segmentation and recognition based on HMM

Abstract

Gesture recognition is a big area of artificial intelligence, gesture segmentation is the difficult problem of continuous vocabulary gesture recognition. There are many automatic techniques to segment gesture, however, most of them have an time interval between the gesture segmentation and output recognition results. The interval is not great for performance of continuous gesture recognition. In order to avoid the time interval, a novel method of continuous vocabulary gesture recognition is proposed. In our method, the start point and the end position of every gesture sequence are found by judging the change of the probability. The probability is the probability of gesture sequence occurrence that is defined by the gesture sequence in the Hidden Markov Model (HMM). We also propose a method to automatically determine the threshold used in the algorithm, which can effectively improve the segmentation accuracy and make the algorithm having better robustness. In the experiment, 93.88 % accuracy can be obtained to the gesture segmentation and 92.22 % accuracy can be obtained to the gesture recognition after segmented.