Identification Detection and YoloV5 based Driver Drowsiness Framework

Human drivers bring a unique mix of skills, instincts, and emotions to the road, shaped by their individual driving habits. However, driver drowsiness poses a serious threat to road safety, making it critical to develop effective detection systems to prevent accidents. Past efforts to identify unusual driver behavior often focused on analyzing the driver’s face or vehicle movements using computer vision techniques. While these methods provided some insights, they struggled to capture the full complexity of driver behavior. With the rise of deep learning, researchers have increasingly turned to neural networks to better understand and detect driver drowsiness. This paper presents a fresh approach using vision transformers and YOLOv5 architectures to recognize signs of drowsiness. We propose a customized YOLOv5 model, pre-trained to detect and extract the driver’s face as the Region of Interest (ROI). To overcome the limitations of earlier systems, we incorporate vision transformers for binary image classification. Our model was trained and tested on the public UTA-RLDD dataset, achieving impressive results with 96.2% training accuracy and 97.4% validation accuracy. To further evaluate its performance, we tested the framework on a custom dataset of 39 participants under various lighting conditions, where it achieved a solid 95.5% accuracy. These experiments highlight the strong potential of our approach for real-world use in smart transportation systems, paving the way for safer roads.