A Dempster-Shafer Evidence Theory Based Multimodal Human Gesture Recognition Method

Hand gestures are widely used in human-computer interaction, and dynamic gesture recognition is still a challenging task. In this paper, a Dempster-Shafer evidence theory based multimodal human gesture recognition method is proposed. Firstly, the audio-based and skeleton-based command recognition models are established. Then, an alignment method for multimodal recognition results of continuous gestures is proposed to combine the recognition results of the audio-based model and the skeleton-based model for the same action into the same group. Furthermore, for the results in each group, the Dempster-Shafer evidence theory is used for fusion. Finally, the performance of our method is evaluated using the ChaLearn Multi-modal Gesture Recognition dataset. The results show that this method can effectively improve the recognition accuracy of dynamic gestures by fusing information from audio and skeleton.