Drowsy driving is an important cause of road accidents that can lead to many fatalities and monetary losses. Moreover, in the upcoming SAE level 3 (conditionally automated driving), the state of the drivers must be monitored since the driver must be attentive to drive manually when the automated driving system cannot control the car any more. To detect driver drowsiness, three data sources have been generally used in the literature: vehicle-based data, facial-based data, and biosignals. Recent studies mostly focused on designing driver drowsiness detection systems using binary classifiers that report the driver’s vigilance into two classes, alert and drowsy. However, adding a middle level of drowsiness can help better estimate the transition between alertness and drowsiness to warn the driver early enough to prevent impaired driving. In addition, previous works mainly concentrated on driver drowsiness detection in manual driving mode, whereas there is no input from the driver in SAE level 3 automated driving. Therefore, the drowsiness detection system cannot utilize vehicle-based data to estimate the drowsiness in automated driving. To address the issues of the previous works, this thesis proposes three new approaches to classify driver drowsiness in simulated driving tests.weiterlesen