Multi-head graph attention network for human pose recognition
Yingyin Fan, Jie Qing Tan, Wanyi Li, Jialing Wu · 2025
Human pose recognition can classify actions by modeling spatial and temporal relationships between body joints, often represented as graph-structured data. To address challenges in capturing dependencies and handling noise, this study proposes the Multi-Head Graph Attention Network (MHGAT). MHGAT combines multi-head attention with temporal modeling to adaptively focus on critical joints and motion dynamics. Validated on benchmark datasets, it outperforms existing methods, advancing pose recognition and its applications in behavior analysis, human-computer interaction, and metaverse development.