Dynamical-Constrained Graph Convolutional Networks for Human Motion Prediction

In the field of human motion prediction, graph convolutional network (GCN)-based methods have demonstrated remarkable performance. To further enhance prediction accuracy, this paper proposes an improved GCN architecture that integrates principles of dynamics. The model introduces dynamics prediction modules to achieve more realistic motion distributions in critical frame forecasting. Experimental evaluations on Human3.6M and AMASS datasets confirm that the proposed approach significantly enhances motion prediction performance.