Association of Visual-Based Signals with Electroencephalography Patterns in Enhancing the Drowsiness Detection in...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Human-Computer Interaction

Background:

• Obstructive sleep apnea (OSA) significantly increases accident risk due to excessive daytime sleepiness.
• Current drowsiness detection methods like PERCLOS face limitations due to environmental factors and require physiological validation.
• Accurate drowsiness detection is vital for assessing fitness to drive in OSA patients.

Purpose of the Study:

• To develop and validate a visual-based drowsiness detection system for OSA drivers.
• To correlate visual drowsiness metrics with electroencephalography (EEG) signals.
• To assess the feasibility of using limited EEG channels for drowsiness detection.

Main Methods:

• Employed adaptive thresholding and eye aspect ratio analysis using OpenCV and Dlib for visual drowsiness detection from video.
• Recorded six-channel EEG data from 50 OSA drivers during a 50-minute driving simulation.
• Utilized discrete wavelet transform to extract EEG features and correlated them with visually identified drowsiness and wakefulness episodes.

Main Results:

• The visual-based method successfully identified numerous drowsiness and wakefulness episodes.
• Theta-alpha-ratio derived from EEG showed a high correlation (94.7%) with visual drowsiness scoring.
• Frontal brain regions, particularly channel F4, demonstrated strong alignment with EEG-based drowsiness detection.

Conclusions:

• Visual-based drowsiness detection shows promise for real-time application in OSA drivers.
• Specific EEG ratios (theta-alpha, delta-alpha, delta-theta) effectively map to visual drowsiness metrics.
• Targeting frontal or occipital EEG channels could reduce hardware requirements for reliable drowsiness monitoring.