Spatial-Velocity Dual-Channel Fusion Based Progressive Graph Convolutional Network for 3D Human Motion Prediction

3D human motion prediction is a core task in virtual reality, human-computer interaction and related fields, with its key challenge lying in modeling complex spatiotemporal dependencies among human joints. To address the limitations of existing graph convolution methods in dynamic feature fusion and node importance differentiation, this paper proposes a Spatial-Velocity Dual-Channel Fusion based Progressive Graph Convolutional Network (SVPGCN). The model independently processes spatial position and motion velocity information while designing a dynamic adjacency matrix interaction mechanism to achieve collaborative optimization between static topological constraints and dynamic motion features. Furthermore, a progressive target-guided learning strategy is introduced to gradually refine prediction results through multi-stage intermediate supervision, effectively mitigating error accumulation in long-term prediction. Extensive experiments demonstrate superior performance over state-of-the-art methods on both Human3.6M and CMU-Mocap datasets. Ablation studies further validate the effectiveness.