Application of Gesture Recognition Based on Spatiotemporal Graph Convolution Network in Virtual Reality Interaction

Abstract

Aiming at the low recognition rate of traditional gesture, a gesture recognition algorithm based on spatiotemporal graph convolution network is proposed in this paper. Firstly, the dynamic gesture data were preprocessed, including removing invalid gesture frames, completing gesture frame data, and normalization of joint length. Then, the key frame of the gesture is extracted according to the given coordinate information of the hand joint. A connected graph is constructed according to the natural connection of time series information and gesture skeleton. A spatio-temporal convolutional network with multi-attention mechanism is used to learn spatio-temporal features to predict gestures. Finally, experiments are carried out on 14 types of gesture datasets in DHG-14 dynamic gesture dataset. Experimental results show that this method can recognize gestures accurately.