Driver fatigue transition prediction in highly automated driving using physiological features. (1st June 2020)

• Record Type: Journal Article
• Authors: Zhou, Feng; Alsaid, Areen; Blommer, Mike; Curry, Reates; Swaminathan, Radhakrishnan; Kochhar, Dev; Talamonti, Walter; Tijerina, Louis; Lei, Baiying
• Is Part Of: Expert systems with applications. Volume 147(2020)
• Publication Date: 2020-06-01
• Subjects: Driver fatigue -- PERCLOS -- Fatigue transition prediction -- Highly automated driving; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.113204
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: We predict the transition from non-fatigue to fatigue using physiological features. We capitalize on PERCLOS as the ground truth of driver fatigue. We select most critical physiological features to predict driver fatigue proactively. We predict the fatigue transition using nonlinear autoregressive exogenous network. The accuracy of fatigue transition prediction evidences the potential of our method. Abstract: One of the main causes of traffic accidents is driver fatigue due to monotonous driving, sleep deprivation, boredom, or a combination of these. Thus, fatigue detection systems have been proposed to alert drivers. However, how early driver fatigue can be detected often determines the effectiveness of the system. Traditional approaches aim to detect driver fatigue in real time, which can be too late in many critical situations, such as the takeover transition period in highly automated driving. Therefore, in this research, we aim to predict the driver's transition from non-fatigue to fatigue in highly automated driving using physiological features. First, we capitalized on PERCLOS (i.e., PERcent of time the eyelids CLOSure) as the ground truth of driver fatigue. Next, we selected the most important physiological features to predict driver fatigue proactively. Finally, using these critical physiological features, we built prediction models that were able to predict the fatigue transition at least 13.8 s ahead of time using a technique called nonlinear autoregressive exogenous network. The accuracy of fatigue transition prediction was promising for highly automated driving ( F 1 measure = 97.4% and 99.1% for two types of models), which demonstrated the potential of the proposed method.