Multi-anchor adaptive fusion and bi-focus attention for enhanced gait-based emotion recognition

Area of Science:

• Computer Vision
• Artificial Intelligence
• Behavioral Science

Background:

• Gait-based emotion recognition shows potential for public safety and healthcare.
• Existing methods struggle with feature redundancy and temporal dynamics.

Purpose of the Study:

• To develop a novel temporal graph convolutional network (MDT-GCN) for enhanced gait-based emotion recognition.
• To address limitations of existing methods by incorporating multi-anchor and bi-focus attention mechanisms.

Main Methods:

• Utilized graph convolutional networks (GCN) and temporal convolutional networks (TCN) to extract pose and action features.
• Integrated multi-anchor attention (MAAF) for multi-scale temporal feature extraction.
• Employed bi-focus attention (BFA) to capture both local and global features.

Main Results:

• MDT-GCN achieved 90.11% accuracy on the Emotion Gait dataset.
• The model reached 84.23% accuracy on the Emotion Walk dataset.
• Demonstrated superior performance over existing methods in emotion recognition from gait.

Conclusions:

• The proposed MDT-GCN effectively recognizes emotions from gait by capturing complex temporal dynamics.
• Open-sourcing code and datasets will foster further research in this interdisciplinary field.