An Intelligent Vision Transformer-Based Real-Time Driver Drowsiness Detection System for Ac-curate Eye State Classification and Road Safety Enhancement
Authors-S.Venkateswara Rao, Marupakula Shireesha
Keyword-Driver Drowsiness Detection, Vision Transformer (ViT), Deep Learning, Transfer Learning, Computer Vision, Eye State Classification, OpenCV, Haar Cascade, Image Processing, Real-Time Monitoring, Driver Fatigue, Road Safety, Artificial Intelligence.
Driver drowsiness is a major factor contributing to road accidents and poses a serious threat to public safety. Continuous monitoring of a driver's alertness can significantly reduce fatigue-related accidents by providing timely warnings. This research proposes a real-time driver drows-iness detection system based on a Vision Transformer (ViT) model for accurate eye state classi-fication. The system utilizes a dataset containing approximately 84,900 images of open and closed eyes collected under different lighting and environmental conditions. Before training, the dataset undergoes preprocessing, label encoding, class balancing through random oversampling, and image augmentation techniques such as rotation, cropping, resizing, and sharpness adjust-ment to improve model robustness. The proposed framework employs the pre-trained Vision Transformer (ViT) model (google/vit-base-patch16-224-in21k) with transfer learning and addi-tional fully connected layers to enhance feature extraction and classification performance. The dataset is divided into 80% training, 10% validation, and 10% testing for effective model evalua-tion. The trained model is integrated with OpenCV and Haar Cascade face detection to perform real-time eye state recognition using a webcam. Whenever continuous eye closure is detected, an alarm is generated to alert the driver and prevent potential accidents. Experimental evaluation is performed using Accuracy, Precision, Recall, and F1-Score. The proposed system achieves an overall 98.8% accuracy, demonstrating its effectiveness in accurately identifying driver drowsi-ness while maintaining reliable real-time performance. The developed framework provides a practical, intelligent, and efficient solution for enhancing driver safety and reducing fatigue-related road accidents.
Doi-[https://doi.org/10.5281/zenodo.21721863]