Citation

Nam DH, Kim GP, Baek KH, Lee DS, Lee HY, Suh MW. Int. J. Automot. Technol. 2022; 23(4): 917-926.

Copyright

(Copyright © 2022, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s12239-022-0080-4

PMID

unavailable

Abstract

Use of an Automated Driving System is expected to improve traffic safety by protecting drivers from drowsy driving. Previous studies on the use of Automated Driving Systems mainly focused on detecting a driver's level of drowsiness and protecting drivers from accidents by performing fallback maneuvers. However, maneuvers conducted in drowsy states are limited in their ability to achieve Minimal Risk Conditions because human drivers show a gradual degradation in their driving ability as they fall asleep and the probability of an accident increases greatly after a driver becomes drowsy. Thus, current Automated Driving Systems require algorithms to predict drowsiness and perform maneuvers before the driver becomes too drowsy. This paper suggests an algorithm that not only detects but also predicts driver drowsiness using 6 vehicle data points. Driver condition is classified into 4 states and Driver drowsiness can be predicted by detecting the severe fatigue state, which tends to occur one minute before the drowsy state. The vehicle driving data are collected using a simulator and features that can be used to distinguish between the 4 states are investigated through data analysis. Ultimately, an optimum machine learning algorithm that can predict driver drowsiness is developed using the investigated factors.

Language: en

Keywords

Driver drowsiness; Light fatigue; Machine learning algorithms; Minimal risk condition; Severe fatigue; Vehicle driving data