Driver Cognitive Workload Detection via Eye-tracking and Physiological Modalities

Abstract

In recent years, autonomous car development has become one of the hottest topics in AI applications and the driver cognitive workload monitoring system is a critical element of the autonomous car. This study explored the feasibility of classifying driver cognitive workload levels with eye-tracking and physiological modalities individually. Around a 70% detection accuracy was obtained with both modalities for ternary classes. Support Vector Machines (SVM) with a Gaussian Kernel function are utilized to build a monitoring system with 5-fold cross-validation. Principal component analysis (PCA) was investigated in terms of system performance. The time gaps between training and testing data are analyzed and the feasibility of using the o-line pretrained model to detect driver cognitive workload is investigated.