Preliminary analysis of full-scale driving simulator data for unmasked sleepiness detection

A driver can mask his sleepiness. This study aims to determine effective and reliable indications of a driver's unmasked sleepiness using driver-vehicle data. A Bayesian approach and the signal detection theory were applied to investigate the effectiveness of selected driver-vehicle parameters for this purpose. Twenty subjects participated in three consecutive driving sessions on the simulated 4-lane highway from Seoul to Cheonan, Korea, during which their PERCLOS (percentage of eye closure) data, assumed to be a true indicator of a driver's unmasked sleepiness, i.e., drowsiness, were monitored. Correlations between PERCLOS and the selected vehicle parameters, such as velocity RMSE (root-mean-square error), were analyzed while participants performed skill-based and rule-based driving tasks. The preliminary experimental results demonstrated that unmasked sleepiness, as indicated by PERCLOS, was not correlated with the selected vehicle parameters for skill-based tasks. Some rule-based tasks, such as VPVT (Visual Psychomotor Vigilance Task), showed significant correlations with masked and unmasked sleepiness, which shows that driver-vehicle data can potentially be used as a dynamic unmasked sleepiness indicator. More in-depth analysis is being conducted and is expected to be included in the final version of the manuscript.