Vision Transformer-Based Real-Time Driver Drowsiness Monitoring with Enhanced Safety Performance

Jithendra Reddy Dandu Department ofElectronics and Communication
Engineering Annamacharya Institute ofTechnology and Science
Tirupati,Indiajithendrareddy.d@gmail.com
Chippagiri Mahesh Reddy
Department of Electronics andcommunication Engineering
Annamacharya institute of technologyand Sciences
Tirupati,Indiamaheshreddyc072@gmail.com
Poola Jyothika Reddy
Department of Electronics andcommunication Engieering
Annamacharya institute of technologyand Sciences
Tirupati,IndiaJyothikareddy345@gmail.com
Anduri Manisha Reddy
Department of Electronics andcommunication Engineering
Annamacharya institute of technologyand Sciences
Tirupati,Indiamanireddy092200@gmail.com

Yadadhala Kalyan Reddy
Department of Electronics andcommunication Engineering
Annamacharya institute of technologyand Sciences
Tirupati,Indiaykalyanreddy.8801@gmail.com

ABSTRACT-Fatigue and drowsiness of drivers are one of the primary reasons for road accidents across the world. Driver monitoring systemsare essential for road safety. This research proposes a real-time driver tiredness detection system using a fine-tuned Vision Transformer. Theproposed system focuses on identifying eye states (open or closed) to determine the level of driver alertness and provide timely warnings toprevent accidents. A Vision Transformer architecture was fine-tuned by adding additional layers and training it on a large dataset consistingof 84,900 images representing open-eye and closed-eye conditions. The transformer-based model effectively captures spatial features andattention patterns in eye images, enabling accurate classification of driver alertness states. The system operates in real time using a camerathat continuously monitors the driver's face and analyzes eye movements. When the model detects prolonged eye closure indicating possible drowsiness, an alarm mechanism is activated to alert the driver immediately. This real-time warning helps the driver regain attention and reduces the likelihood of fatigue-related accidents. The experimental results show that the proposed system has an accuracy of 98.8% along with high precision, recall, and F1-score for both open eye and closed eye detection. This demonstrates the results obtained by employing Vision Transformer-based models for improving the performance of the driver monitoring system significantly. This presented system for real-time monitoring along with the application of sophisticated deep learning (DL) models provides a favourable solution for improving the efficiency of the driver monitoring system.