Abstract

The rising global prevalence of vehicles has led to an escalation in traffic accidents, emerging as a major contributor to human fatalities. This research focuses on improving traffic safety by addressing the issue of drowsy driving. It suggests an algorithm for detecting driver drowsiness in real-time, considering the unique differences among individual drivers. The algorithm uses a Deep Cascaded Convolutional Neural Network (DCCNN) to detect the driver's face, including landmarks with the help of the Dlib toolkit. A novel metric termed "Eye Aspect Ratio (EAR)" is introduced to assess drowsiness by analyzing the landmarks of the driver's eyes. The algorithm consists of two modules: offline training, where a fatigue state classifier is trained, and online monitoring, where the classifier is applied to detect drowsiness quickly from driver images. It employs a variable determined by the frequency of drowsy frames per unit of time for drowsiness evaluation. Comparative experiments indicate that this algorithm surpasses current drowsiness detection methods in terms of both accuracy and speed. The findings from this research have the potential to enhance intelligent transportation systems, improve driver safety, and mitigate losses associated with drowsy driving.