Multi-modal Adversarial Regressive Transformer for Cross-subject Fatigue Detection

Area of Science:

• Neuroscience
• Traffic Safety Engineering
• Machine Learning

Background:

• Fatigue driving is a significant cause of traffic accidents.
• Accurate fatigue detection systems are essential for enhancing road safety.

Purpose of the Study:

• To develop an advanced fatigue detection model using multi-modal data.
• To introduce the Multi-modal Adversarial Regressive Transformer (MART) for fatigue monitoring.

Main Methods:

• Utilized Electroencephalogram (EEG) and Electrooculogram (EOG) data.
• Developed the MART model, incorporating Transformer architecture for multi-modal processing.
• Employed adversarial domain generalization to reduce individual differences.

Main Results:

• The MART model achieved a Root Mean Square Error (RMSE) of 0.1015 in a subject-dependent setup.
• In a cross-subject setup, the RMSE was 0.1556, reduced to 0.1249 with adversarial domain generalization.
• Demonstrated the model's efficacy and adaptability in fatigue monitoring across different subjects.

Conclusions:

• The proposed MART model shows significant promise for real-time fatigue detection.
• Adversarial domain generalization enhances the model's robustness and applicability in diverse scenarios.
• Multi-modal data fusion with advanced Transformer architectures offers a powerful approach to fatigue monitoring.