The Combined Effect of Alcohol and Sleepiness on Eye Tracking Features: Implications for Driver Monitoring Systems

Camera-based Driver Monitoring Systems (DMS) utilise eye tracking metrics to detect impaired behaviours in drivers. This study examines how gaze, blink, and saccade metrics are influenced by alcohol consumption and sleepiness, as well as the combined effects of these impairments on eye tracking features. Seventy-six participants drove on a test track under two conditions: daytime (alert) and night-time (sleep-deprived), with varying alcohol levels up to 1.2‰. Mixed-effects analyses of covariance revealed that as intoxication and sleepiness levels increased, eye blinks became longer, and saccades became fewer, slower, and exhibited lower amplitude. Increased sleepiness, but not increased alcohol levels, was associated with reduced visual scanning and decreased gaze entropy. Many features demonstrated an additive effect of sleepiness and intoxication. Since most of the eye tracking features are similarly affected by both sleepiness and alcohol intoxication, distinguishing between these impairments is challenging. A support vector machine classifier achieved an accuracy of only 67.6% in the 2×2 task of distinguishing between sober or drunk and alert or sleepy drivers. This suggests that DMSs targeting multiple driver states may require additional sensors or modalities to effectively differentiate between various impairments.