A Holistic Approach to Detecting Distracted and Drowsy Drivers

Distracted and drowsy driving is a significant cause of road accidents worldwide. This paper presents a novel system for detecting distracted and drowsy drivers using computer vision and machine learning techniques. This paper leverages multi-YOLOv8 models specifically trained to detect distinct distracted behaviors: mobile phone usage, yawning and seatbelt, achieving an accuracy of 95 %. By focusing on these critical indicators of driver inattention, our system achieves precise, real-time monitoring. The system also calculates the eye-aspect ratio and estimates head pose to assess further signs of drowsiness and inattention. An alertness meter, updated based on these factors, triggers an alarm when the alertness meter exceeds a certain threshold. The models were trained on a combination of a Kaggle dataset and a custom dataset, which includes instances of yawning, mobile phone usage, seatbelt usage, smoking, and drinking. This multi-model approach not only enhances detection accuracy but also addresses various distracted driving behaviors simultaneously, marking a significant advancement over traditional single-model and sensor-based systems.