Objectives: Drowsy driving is a serious highway safety problem. If drivers could be warned before they became too drowsy to drive safely, some drowsiness related crashes could be prevented. The presentation of timely warnings, however, depends on reliable detection. To date, the effectiveness of drowsiness detection methods has been limited by their failure to consider individual differences. The present study sought to develop a drowsiness detection model that accommodates the varying individual effects of drowsiness on driving performance.
Methods: Nineteen driving behavior variables and four eye feature variables were measured as participants drove a fixed road course in a high fidelity motion-based driving simulator after having worked an 8-h night shift. During the test, participants were asked to report their drowsiness level using the Karolinska Sleepiness Scale at the midpoint of each of the six rounds through the road course. A multilevel ordered logit (MOL) model, an ordered logit model, and an artificial neural network model were used to determine drowsiness.
Results: The MOL had the highest drowsiness detection accuracy, which shows that consideration of individual differences improves the models’ ability to detect drowsiness. According to the results, percentage of eyelid closure, average pupil diameter, standard deviation of lateral position and steering wheel reversals was the most important of the 23 variables.
Conclusion: The consideration of individual differences on a drowsiness detection model would increase the accuracy of the model’s detection accuracy.
Xuesong Wang, Chuan Xu. Driver drowsiness detection based on non-intrusive metrics considering individual specifics. Accident Analysis & Prevention, Volume 95, Part B, October 2016, Pages 350–357.