A Hierarchical Attention GCN Network for Body-Hand Gesture Recognition

In this paper, we propose a hierarchical attention GCN network for body-hand gesture recognition that consists of three components. Part A: Body-Hand Skeleton Capture, which captures both body and hand skeleton data simultaneously; Part B: Spatial GCN for Hand Gesture, which recognizes static hand gestures, to obtain the motion parameters of the control; Part C: Multi-Attention GCN is embedded with three dimensions: channel, time, and graph spatial, respectively, to accurately identify dynamic body gestures and obtain the motion mode of the control. The information obtained from parts A and B are combined to obtain the complete control command. The final rich experiments were conducted on two public datasets, which showed that our proposed method achieves effective control of the UAV with significant advantages over advanced methods.