Drivers' Distraction Detection Using Machine Learning in Smart Computing Era

The rising number of road accidents globally has become a major public concern, with driver distraction identified as a key contributor to collisions causing serious injuries, property damage, and fatalities. In this context, monitoring and analyzing driver behavior to detect distractions and reduce accidents is essential. This research presents a deep learning-based method for detecting distracted drivers and identifying distraction sources, such as talking, drinking or adjusting the radio. The study explores several pre-trained models and evaluates them using a dataset of real-world images depicting distracted driving behavior. Standard performance metrics were employed to measure the model's effectiveness. The novelty of this work lies in its ability to recognize various types of distractions, offering valuable insights for road safety authorities to develop targeted policies and strategies aimed at minimizing accident rates caused by distracted driving. The proposed model leverages YOLO-v8, a CNN-based framework, with fine-tuned hyperparameters (e.g., mosaic=0.5, mixup=0.3), and achieved an impressive accuracy of 97.8%. The outcomes of this study hold significant promise for improving road safety and decreasing fatalities and material losses due to distraction-related accidents. Ultimately, the success of this approach can facilitate the development of advanced driver-assistance systems and support the enforcement of stricter traffic safety regulations.