Detecting Driver Sleepiness From Physiological Indicators Using a CNN-LSTM Self-Attention Model.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Transportation Safety

Background:

• Driver sleepiness is a major cause of road accidents.
• Electroencephalography (EEG) and Electrooculography (EOG) signals exhibit characteristic changes during dozing.
• Dozing states can be classified into onset, duration, and end states based on physiological indicators.

Purpose of the Study:

• To develop a real-time framework for detecting driver sleepiness.
• To refine the classification of dozing sub-states using physiological signal analysis.
• To improve road safety through accurate and timely sleepiness detection.

Main Methods:

• Proposed a framework integrating three Convolutional Neural Network-Long Short-Term Memory-Self-Attention (CLSA) models.
• Combined CNN for local feature extraction with self-attention for global context.
• Evaluated performance on continuous test data from 12 subjects using three-channel signal processing.

Main Results:

• The CLSA framework accurately detected alpha waves and rising edge waveforms for identifying the onset of dozing (Alpha Wave Epoch - AWE).
• The duration sub-state was characterized by the sustained presence of alpha waves.
• The end state was classified into two phenomena (alpha blocking or alpha wave attenuation-disappearance) with high accuracy.

Conclusions:

• The proposed framework enables accurate real-time detection of the onset, duration, and end states of driver sleepiness.
• This approach holds promise for enhancing safety in real-world driving scenarios.
• Accurate identification of sleepiness levels can help mitigate traffic accidents.