Abnormal Driver Behavior Detection Using Deep Learning
Faiz Ali, Mouza Jamal Alnuaimi, Sara Ali Alawadhi, Saeed Abdallah · 2024
Many road traffic accidents occur due to drivers’ reckless behaviors, leading to inattentiveness and unawareness of vehicles or pedestrians on the road. Such unsafe driving can result in legal, financial, emotional, and physical consequences. To prevent these accidents and injuries, drivers should follow traffic rules and avoid distractions. This project aims to reduce traffic accidents by detecting common abnormal driver behaviors and hazards, such as not wearing a seatbelt, using a mobile phone, smoking, eating, and fatigue. The proposed system employs a deep learning-based approach to ensure accuracy, reliability, and efficiency. Utilizing You Only Look Once (YOLO), a real-time object detection algorithm, the system captures and analyzes real-time footage of the driver through a camera connected to a Raspberry Pi. The detection accuracy is measured using intersection over union (IoU), which compares the predicted bounding box with the ground truth. When unsafe behavior is detected, the system alerts the driver and provides verbal guidance for correction. The project will present a physical prototype that captures real-time images of the driver and compares them with a trained dataset.