GLM: A Model Based on Global-Local Joint Learning for Emotion Recognition from Gaits Using Dual-Stream Network

Gait is a distinctive human feature that can be recognized from a distance and has been widely utilized in the field of emotion recognition. In this study, we propose a novel dual-stream model (GLM) for gait emotion recognition that combines the strengths of global and local features. We extract skeleton point gait data from walking videos and process them into suitable inputs for two channels of feature extraction networks, which respectively capture global and local characteristics. To enhance the features and improve recognition accuracy, we further introduce an attention-based feature fusion module. Through experiments on benchmark datasets, our proposed model achieves high accuracy in recognizing emotions from gait data.