Cognition-Guided Complex-Valued Graph Convolutional Network for Gait Emotion Recognition

Gait emotion recognition has played a significant role in the development of intelligent systems. Most existing methods primarily focus on capturing the spatial-local relationships of gait while paying insufficient attention to the temporal features of gait, despite the fact that temporal information contains critical cues for emotion recognition. To address the issue, in this article, a novel model, termed CGCV-GCN, is proposed for gait emotion recognition, which targets at improving the GCN network by integrating complex-valued representation and cognitive hierarchy. Specifically, a Complex-valued Spatial-Temporal Graph Convolutional Network (CST-GCN) is first devised for learning gait emotion representation in a magnitude-phase complex space, which leverages phase information to improve gait emotion features. Furthermore, a cognition-inspired fusion mechanism (CGFM) is proposed to adaptively integrate cross-representation-space and cross-modal features, enabling a multi-level local-global gait feature learning process. Experimental results on public datasets demonstrate that CGCV-GCN significantly improves emotion recognition accuracy over state-of-the-art methods. The code is available at: https://github.com/xiejiachun1902/CGCV-GCN