mmSkeleton: 3D Human Skeleton Estimation Using Millimeter Wave Radar Sparse Point Clouds

Human body pose estimation based on millimeter-wave radar sparse point clouds is an emerging research field. In this paper, we propose mmSkeleton, a network based on spatio-temporal self-attention mechanism, to extract and integrate information embedded in real point clouds. Additionally, we investigate the generalization issue of the network to different actions and propose a point clouds simulation method based on electromagnetic propagation theory and human electromagnetic scattering characteristics. Experimental results demonstrate that mmSkeleton achieves state-of-the-art (SOTA) performance on our own dataset and two publicly available datasets, and the proposed data augmentation method effectively enhances the model's generalization to different actions.