EPIC: Emotion Perception by Spatio-Temporal Interaction Context of Gait

Area of Science:

• Psychophysiological computing
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Gait-based emotion recognition is a growing field within psychophysiological computing.
• Existing methods often overlook spatio-temporal gait contexts, limiting emotion-gait relationship understanding.
• Advanced analysis of joint dynamics is crucial for accurate emotion inference.

Purpose of the Study:

• To propose an integrated emotion perception framework (EPIC) for enhanced gait-based emotion recognition.
• To explore spatio-temporal gait contexts for higher-order emotion-gait relationship capture.
• To develop a novel method for generating synthetic gaits and classifying emotions.

Main Methods:

• Analyzing non-adjacent joint coupling using Phase Lag Index (PLI).
• Synthesizing gait sequences with spatio-temporal constraints using Dynamic Time Warping (DTW) and Gated Recurrent Units (GRU).
• Employing Spatial Temporal Graph Convolution Networks (ST-GCN) for emotion classification.

Main Results:

• The EPIC framework successfully generates synthetic gaits by analyzing spatio-temporal interactions.
• The proposed loss function improves gait sequence accuracy.
• Achieved a classification accuracy of 89.66% on the Emotion-Gait dataset, outperforming existing methods.

Conclusions:

• The EPIC framework offers a novel approach to gait-based emotion recognition by integrating psychophysiological computing and AI.
• Analyzing spatio-temporal gait dynamics significantly improves emotion recognition accuracy.
• This research advances the understanding of the complex relationship between human gait and emotional states.