Detecting Distracted and Drowsy Driving with Deep Learning Techniques and Facial Landmarks

Road safety is a paramount concern globally, and thus Machine Learning and Deep Learning applications have become widespread in this area. Deep Learning models can accurately and efficiently detect potential causes of road accidents. Across numerous years, there have been two primary reasons for road accidents—Distracted and Drowsy driving. Distracted driving means doing anything that diverts a driver’s attention from the road. It includes activities such as texting or talking on the phone, talking to other passengers, and operating the vehicle’s music system. Drowsy driving is when the driver is showing signs of falling asleep, such as yawning, closing eyes or if his head is drooping downwards. This study proposes a model architecture that leverages VGG-16 and Transfer Learning, achieving an accuracy of 97.03% for the identification of distracted driving characteristics. To ensure strong model performance on a wider variety of images, various augmentations have been applied to the images of the training dataset. This paper also introduces an innovative approach to detect driver drowsiness using facial landmarks. The proposed system combines blink and mouth opening ratios alongside head posture to identify signs of driver fatigue, providing a comprehensive solution for improving overall road safety. Such a solution not only enhances the efficiency of accident prevention but also promotes the development of intelligent transportation systems for safer roads.