Detecting Driver Sleepiness

Doctoral thesis, 2011

Driver sleepiness has been identified as one of the most prevalent factors in road crashes. Methods for online detection of driver sleepiness may help to prevent such sleepiness-related crashes. The main aims of this thesis have been to improve driver sleepiness detection and to further the understanding of how sleepiness impairs driving behavior. Stochastic optimization methods have been used for optimizing indicators (measures) of sleepiness (based on driving behavior) found in the literature as well as generalized indicators. Furthermore, (nonlinear) combinations of multiple indicators have also been investigated. In addition to considering indicators based on driving performance, a mathematical model of sleepiness (the Sleep/wake predictor, SWP) has also been included in the nonlinear combinations of indicators. Using data collected in a moving-base driving simulator study, the improvements obtained with stochastic optimization of indicators were modest. In addition, after optimization, the performance scores (taken as the average of sensitivity and specificity, with an upper limit of 1) of different indicators were similar. The best optimized indicator of this kind achieved a score of 0.74. A nonlinear combination of that indicator with SWP gave a performance score of 0.83. By contrast, the SWP alone gave a score of 0.78. Two additional studies were carried out on real roads in actual traffic, involving car and truck driving, respectively. These two studies are of particular importance since most results in the field have been obtained from simulator studies. The question of whether results obtained in a driving simulator can be generalized to real driving is of great importance. In this thesis, a comparison in absolute levels between car driving and simulator driving has been made. The studies involving real driving showed that the average lateral position was indicative of driver sleepiness, whereas changes in lateral variability was not. This was an unexpected and novel finding which contradicts previous findings concerning driving behavior deterioration (due to sleepiness) in simulator studies. It was also found that driving in the moving-base simulator is associated with more severe sleepiness than real driving.