article · Scientific Reports
Driver fatigue contributes substantially to road collisions, creating severe economic and societal harm. To address this, a contactless deep learning system evaluates eye states in real time to detect driver drowsiness. The approach incorporates image normalisation, augmentation, and Haar Cascade classifiers for region-of-interest selection, classifying eye status as open or closed. Across multiple evaluated models, Vision Transformer and Swin Transformer architectures demonstrated the highest performance, achieving accuracy rates of 99.15 per cent and 99.03 per cent respectively on the MRL Eye Dataset. Testing across additional datasets, including NTHU-DDD and CEW, confirmed robustness across diverse conditions such as varying lighting and subjects wearing glasses. The framework integrates a scoring mechanism that triggers an alert during prolonged eye closure, alongside Class Activation Mapping to make model decisions transparent by highlighting critical eye regions during inference.
Driver drowsiness is a major contributor to road traffic accidents, leading to severe injuries, fatalities, and economic loss. Developing reliable, contactless monitoring systems capable of operating accurately under difficult conditions, such as poor lighting or through spectacles, offers a practical way to intervene before fatigue leads to collisions, directly improving safety for drivers, passengers, and other road users.
The abstract points directly towards integration into advanced driver assistance systems to provide real-time, contactless fatigue alerts for motorists and transport operators. Having undergone testing on multiple benchmark datasets under challenging scenarios like changing light levels and eyewear, the technology appears to be applied and tested in simulated or benchmark settings. Moving to commercial deployment would require porting the models into automotive in-cabin monitoring hardware.
AI-generated from the published abstract. Always read the original work before citing.
Driver drowsiness is a leading cause of road accidents, resulting in significant societal, economic, and emotional losses. This paper introduces a novel and robust deep learning-based framework for real-time driver drowsiness detection, leveraging state-of-the-art transformer architectures and transfer learning models to achieve unprecedented accuracy and reliability. The proposed methodology addresses key challenges in drowsiness detection by integrating advanced data preprocessing techniques, including image normalization, augmentation, and region-of-interest selection using Haar Cascade classifiers. We employ the MRL Eye Dataset to classify eye states into "Open-Eyes" and "Close-Eyes," evaluating a range of models, including Vision Transformer (ViT), Swin Transformer, and fine-tuned transfer learning models such as VGG19, DenseNet169, ResNet50V2, InceptionResNetV2, InceptionV3, and MobileNet. The ViT and Swin Transformer models achieved groundbreaking accuracy rates of 99.15% and 99.03%, respectively, outperforming all other models in precision, recall, and F1-score. To ensure the generalization and robustness of the proposed models, we also evaluate their performance on the NTHU-DDD and CEW datasets, which provide diverse real-world scenarios and challenging conditions. This represents a significant advancement over existing methods, demonstrating the effectiveness of transformer-based architectures in capturing complex spatial dependencies and extracting relevant features for drowsiness detection. The proposed system also incorporates a real-time drowsiness scoring mechanism, which triggers alarms when prolonged eye closure is detected, ensuring timely intervention to prevent accidents. A key novelty of this work lies in the integration of Class Activation Mapping (CAM) for enhanced model interpretability, allowing the system to focus on critical eye regions and improve decision-making transparency. The system was rigorously tested under varying lighting conditions and scenarios involving glasses, showcasing its robustness and adaptability for real-world deployment. By combining cutting-edge deep learning techniques with real-time processing capabilities, this research offers a contactless, reliable, and efficient solution for driver drowsiness detection, significantly contributing to improved road safety and accident prevention. The proposed framework sets a new benchmark in drowsiness detection, highlighting its potential for widespread adoption in advanced driver assistance systems.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1038/s41598-025-02111-x
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
