: Driver drowsiness is a major contributor to road accidents worldwide. Fatigue and lack of sleep are common issues among drivers, often resulting in decreased alertness and increased risk of collisions. One effective way to prevent such accidents is by detecting and alerting drivers about their drowsiness in advance. Various methods exist for identifying signs of drowsiness.
In this study, we propose a deep learning-based solution for detecting driver fatigue. Our approach leverages transfer learning using MobileNet, a lightweight convolutional neural network architecture. We utilized a dataset focused on the eye region to accurately identify signs of drowsiness.
Introduction
Falling asleep while driving is a serious safety hazard, with men twice as likely as women to experience drowsiness behind the wheel. Drowsy driving impairs mental function similarly to alcohol, increasing accident risks. In the U.S., about 23% of adults admit to dozing off while driving, and over half drive while drowsy annually.
Literature Survey:
Research focuses on driver behavior, including environmental, psychological, and physiological factors.
Advances in driver monitoring use big data and IoT to detect fatigue and distraction in real time.
Existing Systems:
Previous models using machine learning showed limited accuracy in detecting drowsiness.
Proposed System:
A deep learning approach using a CNN-based MobileNet transfer learning model is developed. Real-time video input is processed via OpenCV, analyzing eye closure frame-by-frame. If eyes stay closed beyond a set threshold, an alert siren warns the driver and passengers.
Modules:
Dataset Storage: User data is stored for model training.
Model Training: Data is preprocessed and used to train the model for accurate fatigue detection.
Prediction: The trained model predicts drowsiness from live video input, focusing on eye and mouth movements.
System Architecture and Data Flow:
Wearable sensors collect physiological signals.
Data is transmitted to a mobile app and backend server.
After preprocessing, deep learning models are trained and evaluated.
The system provides real-time drowsiness detection and alerts through the app.
Results and Analysis:
The system effectively detects driver fatigue by monitoring eye closure and yawning using facial landmarks and head pose estimation, delivering reliable real-time alerts to prevent accidents.
Conclusion
TheConnectivity-Aware Graph Neural Network (CAGNN) provides a powerful and efficient method for detecting driver drowsiness in real time. It combines Graph Neural Networks (GNNs), Gated Recurrent Units (GRUs), and Random Forest classifiers to effectively learn both spatial and temporal features from the data. Its ability to adapt well to real-world driving conditions and maintain strong performance makes it a promising tool for enhancing road safety and minimizing fatigue-related incidents.
References
[1] G. Sikander and S. Anwar, “A review of driver fatigue detection systems,” IEEE Trans. Intell.https://ieeexplore.ieee.org/document/8460900
[2] M. S. Clayton, N. Yeung, and R. Cohen Kadosh, “The role of cortical oscillations in sustained attention,” Trends Cognit. Sci., vol. 19, no. 4, pp. 188–195, Apr. 2015.https://www.sciencedirect.com/science/article/pii/S1364661315000285
[3] L. De Gennaro and M. Ferrara, “An overview of sleep spindles,” Sleep Med. Rev., vol. 7, no. 5,pp.423–440,Oct.2003.https://www.sciencedirect.com/science/article/pii/S1087079202902522
[4] G. Borghini, L. Astolfi, G. Vecchiato, D. Mattia, and F. Babiloni, “Neurophysiological measurements of pilots and drivers to assess mental workload, fatigue, and drowsiness,” Neurosci. Biobehavioral Rev., vol. 44, pp. 58–75, Jul. 2014.https://www.sciencedirect.com/science/article/pii/S0149763412001704
[5] T. Åkerstedt, G. Kecklund, and A. Knutsson, “Manifestations of sleepiness and EEG spectral content during shift work,” Sleep, vol. 14, no. 3, pp. 221–225, May 1991.https://pubmed.ncbi.nlm.nih.gov/1896723/