A Gait Recognition Method Using 3D Joint Motion Estimation Spatiotemporal Modeling and Attention Mechanism

Gait recognition, as a technology for identifying individuals based on their motion patterns, has been widely applied in areas such as security monitoring and behavior analysis. However, traditional gait recognition methods primarily rely on 2D skeleton information, which limits their ability to fully capture the intricate features of human motion. As a result, these methods often struggle with multi-view and diverse motion variations in complex scenarios. To address these challenges, this paper proposes a novel gait analysis approach that integrates 3D joint motion estimation, spatiotemporal feature modeling and attention-based feature selection. Specifically, we first utilize a 3D pose estimation network to map 2D key points to 3D space, generating comprehensive skeleton information across spatial and temporal dimensions. Next, spatiotemporal graph convolutional networks (ST-GCN) are employed to model the skeleton sequence, enabling the extraction of dynamic gait features. Finally, a multi-head attention mechanism is introduced to refine feature selection and enhance recognition accuracy. Experimental results on datasets such as OUMVLP and GREW demonstrate that the proposed method outperforms existing approaches across all metrics, achieving a remarkable average precision of 92.4% on the OUMVLP dataset. Additionally, ablation studies confirm the significant contributions of each module to the overall performance improvement in gait recognition.