Gait Emotion Recognition Using a Bi-modal Deep Neural Network

Abstract

Gait Emotion Recognition is an emerging research domain that focuses on the automatic detection of emotions from a person’s manner of walking. Deep learning-based methodologies have been proven highly effective for computer vision tasks. This paper provides a powerful deep-learning architecture for emotion recognition from gait by introducing the fusion of domain-specific discriminative features with latent deep features. The proposed Bi-Modal Deep Neural Network (BMDNN) combines salient features extracted from a deep neural network with highly-discriminating handcrafted features. The proposed architecture outperforms state-of-the-art methods in all emotional classes on the Edinburgh Locomotion MoCap Dataset.