Tai Chi Chuan (TCC) teaching process requires face-to-face interaction between learners and masters, which restrains the promotion of TCC since skilled masters are rare and manual evaluation is time-consuming. Skeleton-based action recognition models are effective for this issue, while the existing human skeletal structures and detection frameworks are inadequate for this task. To address these challenges, a novel human skeletal structure aligned with the essence of TCC is proposed, along with a dataset with 2400 annotated images covering fundamental TCC movements. Furthermore, a module named Waist and Limbs Feature Separation (WLFS) is proposed for structured modeling. Based on the spatiotemporal characteristics of TCC movements, the WLFS module explicitly separates keypoints into dynamic and static categories channel-wise. Subsequently, two exclusive GAUs are applied to the static and dynamic regions respectively. This strategy enables the network to learn the distinct features of the separated categories regions, and accelerates the convergence of the network weights during the training process. To preserve the fine-scale features of TCC keypoints in the downsampling, a Multi-Scale Feature Fusion (MSFF) module is integrated into WLFS, which fuses the different spatial resolution feature maps to enhance feature representation of the model at small scales. Experiments on the custom dataset (Tai Chi) and public datasets (MPII, COCO-WholeBody V1.0, Animal Pose, and AP-10K) demonstrate that the proposed method achieves competitive performance and good generalization ability.

In human pose estimation, formulating keypoint localization as a classification task over discretized coordinate grids has proven effective. Essentially, the 2D features of the keypoints are reduced to 1D coordinate representations. This process leads to the loss of spatial constraints among keypoints and increases the difficulty for the model to capture their structural relationships. To address this issue, we propose an enhanced query attention mechanism constrained by bidirectional graphs. The core idea is to establish the topological constraints on the 1D coordinate representations. First, two fundamental connection directions of the skeleton are defined and encoded as a pair of adjacency matrices to enhance the feature interaction capability of the graph convolutional network (GCN). Second, a GCN-guided multi-scale feature fusion framework is designed to effectively combine multi-scale visual features with structural priors, thereby enhancing the representation of keypoint spatial distributions. Finally, a dual-gate module is incorporated into a GCN-guided attention unit to construct a structured query matrix constrained by the bidirectional skeleton graphs, which helps filter out spurious joint interactions and emphasize plausible ones. Extensive experiments on Tai Chi Chuan-Pose, Animal-Pose, AP-10K, MPII, COCO, and COCO-WholeBody datasets demonstrate that the proposed method outperforms existing methods in terms of both accuracy and robustness, particularly in balancing precise local keypoint localization with global pose consistency.