Optimized driver fatigue detection method using multimodal neural networks.

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Transportation Safety

Background:

• Driver fatigue is a major cause of road accidents.
• Current detection methods often lack precision and reliability.
• There is a critical need for advanced systems to monitor driver alertness.

Purpose of the Study:

• To develop and evaluate a novel multimodal neural network for precise driver fatigue detection.
• To investigate the efficacy of integrating physiological and facial data.
• To enhance road safety by providing reliable fatigue monitoring solutions.

Main Methods:

• Utilized the DROZY dataset containing physiological (EEG, ECG) and facial data.
• Developed two multimodal neural network models: feature combination and feature coupled.
• The highlight model employed a sophisticated coupling mechanism where features mutually weighted each other.

Main Results:

• The multimodal feature-coupled model achieved 98.41% accuracy, 98.38% precision, 98.39% recall, and 98.38% F1-score.
• The multimodal feature combination model achieved 94.87% accuracy.
• A majority voting strategy enhanced decision-making robustness.

Conclusions:

• The multimodal feature-coupled model demonstrates superior performance in driver fatigue detection.
• This approach offers a significant advancement for in-vehicle monitoring systems.
• The findings contribute to developing more effective strategies for preventing fatigue-related accidents.