Driver drowsiness detection through a vehicle's active probe action

Drowsy driving is one of the major causes of traffic collisions, injuries, and fatalities. Existing literature primarily detects driver drowsiness by passively monitoring lanes, steering angles, behavioral states, and physiological states. The paper presents an approach towards enabling vehicles to detect driver drowsiness through the vehicle's active probe action actively. To this end, we record and analyze drivers' responses to a slight active left-lane drifting action of the vehicle in a driving simulator. According to drivers' responses, six indicators of drowsiness are extracted and then used to detect driver drowsiness with three recognition methods, i.e., support vector machine, Gaussian kernel density estimation, and back-propagation neural networks, in comparison to traditional monitoring features regarding steering-wheel movement. Experimental results demonstrate that our proposed active probe approach outperforms the traditional monitor methods for driver drowsiness detection with an accuracy of 97.50%, precision of 95%, and specificity of 98.21%. The proposed active driver drowsiness detection could facilitate a new development of active safety systems.