Real-Time Driver Drowsiness Detection and Warning System

This study presents a deep learning approach for detecting driver drowsiness in real time. Visual cues such as eye opening, blinking and yawning are classified using CNN models and transfer learning, then passed to an audio warning module.

Scope

The system observes drivers through a camera to detect closed eyes, blinking patterns and yawning, aiming to reduce accident risk through timely warnings.

Data and preprocessing

• An open eye dataset with approximately 85,000 images.
• A yawning dataset with more than 5,000 images.
• Class balancing, augmentation and noise-resistant preprocessing.

Method

Eye and mouth localization is followed by feature extraction and binary or multiclass classification. A lightweight model supports real-time operation.

Reported results

• Accuracy: 98.0%.
• Sensitivity/recall: 97.5%.
• Specificity: 98.2%.

Contributions and future work

The study combines contactless visual detection, a balanced training pipeline and immediate warnings. Future work includes additional behavioral cues, broader field testing and integration with fleet management systems.