Advancements in Drowsiness Detection: A YOLOv5 Approach

Drowsiness and fatigue present significant safety and productivity risks across various domains. This study proposes a novel computer vision-based approach for real-time drowsiness detection using the YOLOv5 object detection algorithm. The system was trained on a custom dataset of individuals in awake and drowsy states, achieving an exceptional accuracy of $\mathbf{9 8 \%}$ in classifying signs of sleepiness based on facial features and eye movements captured through webcam feeds. The integration of YOLOv5 with OpenCV and PyTorch enables efficient processing of video frames for continuous real-time monitoring of alertness levels. Comparative analysis with recent YOLOv5-based drowsiness detection studies shows our system’s $98 \%$ accuracy outperforms similar applications. Quantitative comparison of YOLOv5 variants revealed mAP ranging from $\mathbf{2 8. 0 \%}$ (YOLOv5n) to $\mathbf{4 9. 0 \%}$ (YOLOv51) at IoU 0.5:0.95, demonstrating trade-offs between model size, accuracy, and computational complexity. Compared to traditional methods, this vision-based approach offers a more practical and user-friendly solution for drowsiness detection, with potential applications in transportation, industrial automation, healthcare, and security. Future research directions include enhancing accuracy, exploring multi-modal sensing, and addressing data privacy concerns, aiming to advance drowsiness detection technology.