| With growing popularities of intelligent in-vehicle technologies, monitoring of driver cognitive load is becoming more important for both safety and comfort concerns. This task is recognized to be challenging due to limited prior knowledge. This thesis explores the feasibility of classifying driver cognitive load levels based on visual attention features.;First, we contribute a dataset collected from 37 experienced drivers. The collection process focuses on eliciting three levels of cognitive load effectively. The resulting dataset consists a total of eight measurements, gathered from visual, vehicular, physiological sensors and subjective-questionnaires.;Next, we focus on the eye-tracking modality and propose meta-features for capturing variations in visual attention intensity and direction. Then, five machine learning algorithms are applied for subject-independent classification. Issues arose from the machine learning workflow, such as evaluation bias, are examined. The most promising algorithm (Random Forest) achieves 70.3% accuracy on classifying between high and low cognitive load. |
