Transfer Learning-based Driver Distraction Detection

Due to driver’s preoccupation, there have been more accidents on the roads recently all around the world. One of the key explanations is attributed to human behavior. Accidents have been linked mostly to drivers’ distraction. This sudden road crush frequently causes injuries, property loss, or even fatalities. The automated interpretation of the driver’s behavior is one of the trickiest subjects in the field of intelligent transportation systems (ITSs). This study looks on the human action recognition aspect of distracted driving posture recognition. There have been several reports of inattentive driving causing auto accidents. The goal is to use cutting-edge transfer learning techniques like MobileNetV2 and DenseNet to identify distracted driving activities more effectively. As driving is a complex task, it requires complete attention of the driver. Distractions of the driver can be due to talking phone calls, texting, talking to another passenger, drinking, operating radio. To identify distractions while driving, it is crucial to observe and evaluate the driver’s behavior. Images of the driver serve as the main source of data and that comprises of face, arms, and hands of the person inside the car. To decrease vehicle accidents and enhance transportation security, it is extremely desirable and has received a significant amount of study to develop a system that can detect distracted driving. In this study, transfer learning models like DenseNet and MobileNetV2 are used to classify and detect driver distraction. The best performance was produced by MobileNetV2, which has an accuracy of 93.8.