Affective-pose gait: perceiving emotions from gaits with body pose and human affective prior knowledge

As a non-verbal biometric method that can be perceived emotion at a distance, gait has broad applications in affective computing. To perceive emotions from gaits, existing methods usually use velocity, acceleration and area to describe human affective features, which often fail to learn the features of body pose and lose representativeness comprehensively. In this paper, we design the fine-grained affective features based on prior knowledge and present a novel perspective to treat the fine-grained affective features of the gait as the fusion of spatial-temporal features. Following this perspective, we use the ST-GCN to build the pose features and utilize the CNN to learn the affective features. By integrating, we proposed the Affective-Pose Gait network, which fusion the pose and affective feature to analyze the emotions in gaits. The experimental results on the Emotion-Gait dataset prove that Affective-Pose Gait achieves 85.2% in terms of accuracy and outperforms state-of-the-art methods.

Automated summary

This paper presents a novel perspective to treat the affective features of gait as the fusion of spatial-temporal features. By integrating pose and affective features, the proposed Affective-Pose Gait network achieved an accuracy of 85.2% on the Emotion-Gait dataset, outperforming state-of-the-art methods.