A Deep Learning Approach to Automated Drowsiness Detection in Drivers Using CNN Models for Enhanced Road Safety

Traffic accidents are responsible for the majority of fatalities and injuries worldwide. According to the World Health Organization, around one million people die each year due to traffic-related injuries. Drivers who are sleep-deprived, fatigued, or not well-rested may fall asleep behind the wheel, putting both themselves and other road users at risk. Research indicates that drowsiness is a leading cause of major road accidents. In recent times, tired driving has emerged as the primary factor contributing to driver drowsiness, which in turn has led to an increase in road accidents. This has become a critical issue that needs urgent attention. The goal of many technologies is to improve real-time drowsiness detection. Numerous devices have been developed using various artificial intelligence algorithms to address this issue. My research focuses on detecting driver drowsiness by analyzing the driver’s face, tracking eye movement, and alerting the driver with an alarm when drowsiness is detected. In this paper, I utilized various Convolutional Neural Network (CNN) models, including Small CNN, VGG16, VGG19, and Inception, to classify distracted drivers based on the State Farm Distracted Driver Detection challenge on Kaggle. The deep learning framework employed for this task is Keras, which runs on top of TensorFlow. The system compares the extracted eye images with a dataset to detect drowsiness. If the system identifies closed eyes within a certain range, it triggers an alarm to alert the driver. If the driver opens their eyes after the alert, the system resumes tracking. The system adjusts the score based on eye status: the score decreases when eyes are open and increases when eyes are closed. This paper aims to address drowsiness detection with an accuracy of 94%, contributing to the reduction of road accidents.