Li Jincheng, Dai Xuejing, Yan Ruiao, Tang Chengqing, Li Yunpeng
College of Public Security Information Technology and Intelligence, Criminal Investigation Police University of China, Tawan Street83, Shenyang, 110854, Liaoning, China.
College of Frensic Science, Criminal Investigation Police University of China, Tawan Street83, Shenyang, 110854, Liaoning, China.
Sci Rep. 2025 Apr 29;15(1):14946. doi: 10.1038/s41598-025-97922-3.
Gait-based emotion recognition has emerged as a promising field with applications in public safety, healthcare, and human-computer interaction. However, existing methods often suffer from excessive globalization, feature redundancy, and lack of dynamic time dependence. To address these issues, we propose a novel temporal graph convolutional network (MDT-GCN) that integrates multi-anchor (MAAF) and bi-focus attention (BFA) mechanisms. MDT-GCN extracts pose and action features from bone nodes using GCN and TCN networks, respectively. The MAAF module captures multi-scale temporal features to understand emotional expressions across different time ranges, while the BFA module focuses on both local and global features, enhancing the model's ability to capture complex emotional information. Experimental results on the Emotion Gait and Emotion Walk datasets demonstrate the effectiveness of MDT-GCN, achieving recognition accuracies of 90.11% and 84.23%, respectively. By making the code and datasets openly accessible, we aim to facilitate further research and applications in this field. The source code is released on https://github.com/928319204ljc/MDT .
基于步态的情感识别已成为一个有前景的领域，在公共安全、医疗保健和人机交互等方面有应用。然而，现有方法往往存在过度全局化、特征冗余以及缺乏动态时间依赖性等问题。为解决这些问题，我们提出了一种新颖的时态图卷积网络（MDT-GCN），它集成了多锚点（MAAF）和双焦点注意力（BFA）机制。MDT-GCN分别使用GCN和TCN网络从骨骼节点中提取姿势和动作特征。MAAF模块捕捉多尺度时间特征，以理解不同时间范围内的情感表达，而BFA模块关注局部和全局特征，增强了模型捕捉复杂情感信息的能力。在情感步态和情感行走数据集上的实验结果证明了MDT-GCN的有效性，分别实现了90.11%和84.23%的识别准确率。通过公开代码和数据集，我们旨在促进该领域的进一步研究和应用。源代码发布在https://github.com/928319204ljc/MDT 上。
