Multi-scale Spatial Graph Convolutional Network for Skeleton-based Action Recognition

Traditional RGB image action recognition methods are susceptible to environmental occlusion and illumination changes. Although skeleton keypoint data has stronger robustness, existing methods still do not adequately model cross-scale joints. Therefore, this paper proposes a multi-scale graph convolutional network that can simultaneously capture local detail features and global structural information, thereby more efficiently modeling the dependencies between skeleton keypoints at different spatial scales. Extensive experiments are conducted on two large-scale benchmark datasets to verify the effectiveness of the model. The results show that our method outperforms other state-of-the-art methods in terms of accuracy.