Driver Vigilance Monitoring System

Abstract

Driver doziness is a significant contributor to road accidents encyclopedically, challenging robust and real- time discovery systems. This paper presents a comprehensive motorist Alert Monitoring System using deep literacy and computer vision ways to descry motorist fatigue by assaying eye countries. The system employs a Convolutional Neural Network (CNN) trained on grayscale eye images to classify eye countries as\' Open\' or\' Closed\'. MediaPipe\'s Face Mesh result is employed for precise facial corner discovery and birth of eye regions, which are also preprocessed with CLAHE and regularized before feeding into the trained model. A critical aspect of the system involves covering the duration of unrestricted eye countries; if a predefined threshold is exceeded, an audio alarm is touched off to warn the motorist, along with visual cues on the display. The advanced system demonstrates high delicacy in eye state bracket and effective real- time doziness discovery, validated through comprehensive training and evaluation criteria including perfection, recall, F1- score, ROC AUC, and confusion matrices. A stoner-friendly Tkinter- grounded visual interface facilitates system control and provides real- time status updates and access to discovery logs, offering a practical result for enhancing road safety.

Introduction

1. Problem Background

• Driver fatigue is a major cause of road accidents worldwide, impairing reaction time and judgment.

• Traditional solutions like self-assessment or scheduled breaks are often ineffective.

• Vision-based technologies now allow real-time, non-intrusive monitoring of drowsiness using eye behavior, especially eye closure.

2. Proposed System Overview

• A real-time drowsiness detection system was developed using:

  • MediaPipe for facial landmark detection.

  • A custom-built CNN model to classify eye states (Open vs Closed).

  • The PERCLOS metric to detect prolonged eye closure.

  • A webcam-only setup, making it suitable for vehicles.

3. Research Context

• Prior approaches fall into three categories:

  • Physiological (EEG, ECG) – accurate but intrusive.

  • Vehicular (steering/lane analysis) – context-sensitive.

  • Behavioral/Visual – most practical and widely adopted using eye and face analysis.

• Deep learning, especially CNNs, has significantly improved accuracy and reliability in behavioral methods.

4. System Components

A. Data Acquisition & Preprocessing

• Uses webcam feed, flipped and color-converted for processing.

• MediaPipe Face Mesh detects 468 facial landmarks.

• Eye regions are extracted, resized (24×24), converted to grayscale, and enhanced using CLAHE.

• Images are normalized and reshaped for CNN input.

B. CNN Model Design

• Built with Conv2D, ReLU, BatchNorm, MaxPooling, Dropout, and Dense layers.

• Final output uses a sigmoid activation to predict eye state probability.

• Trained using the MRL Eye Dataset with data augmentation and binary cross-entropy loss.

C. Model Training Details

• Image size: 24×24, grayscale.

• Dataset size: ~1,500 labeled images (Open/Closed).

• Uses Adam optimizer, early stopping, and model checkpointing.

• Augmentation includes shifts, rotations, zooms, and brightness changes.

D. Drowsiness Detection Logic

• Eye states are predicted frame-by-frame.

• If eyes are closed for ≥3 consecutive frames and for ≥1.0 seconds, drowsiness is detected.

• A loud alarm is triggered until the eyes are open for at least 4.0 seconds.

• All detection events are logged with timestamps.

5. Key Contributions

• A modular, real-time, deep learning-based drowsiness detection system.

• Fully non-intrusive and designed for practical deployment using only a webcam.

• Improved reliability through MediaPipe + CLAHE + CNN integration.

• Custom logic differentiates between normal blinking and actual fatigue.

Conclusion

This paper presented a robust and real-time Driver Vigilance Monitoring System utilizing deep learning and computer vision for detecting driver drowsiness based on eye state analysis. The system successfully integrates MediaPipe for precise facial landmark detection, a tailored CNN model for eye state classification, and a comprehensive drowsiness detection algorithm with an immediate audio alert system. A thorough assessment revealed the eye state categorization model\'s excellent accuracy as well as the real-time detection and alerting mechanism\'s usefulness. The Tkinter-based Graphical User Interface (GUI) enhances usability, making the system accessible and manageable. The key contributions of this work include: • Development of an efficient CNN model optimized for real-time eye state classification. • Integration of MediaPipe for robust and accurate eye region extraction, combined with custom margin adjustments for improved cropping. • Application of Contrast Limited Adaptive Histogram Equalization (CLAHE) in preprocessing for enhanced contrast in eye images. • Implementation of a multi-threaded system for non-blocking alarm functionality and GUI responsiveness. • A clear, configurable drowsiness detection logic incorporating consecutive frame counts and time thresholds to differentiate blinks from prolonged eye closure. For future work, several enhancements can be explored to further advance the system\'s capabilities: • Integration of Additional Indicators: Combining eye state analysis with other drowsiness indicators such as yawning detection, head pose estimation, or more explicit PERCLOS calculation could further improve the system\'s accuracy and robustness. • Robustness to Lighting Conditions: Further research into advanced image processing techniques or more robust deep learning architectures could enhance performance under extreme or rapidly changing lighting conditions. • Embedded System Deployment: Optimizing the model for deployment on edge devices or embedded systems (e.g., Raspberry Pi, NVIDIA Jetson) could enable standalone, low-cost in-car solutions. Model quantization or pruning techniques could be explored for this purpose. • Driver Identification and Personalization: Developing features for driver identification and personalized drowsiness thresholds based on individual blinking patterns or fatigue levels could enhance user experience and accuracy. • Feedback Mechanisms: Incorporating haptic feedback (e.g., seat vibration) or visual warnings on a dashboard display in addition to audio alerts could provide a more nuanced warning system. • Dataset Expansion: Training the model on a larger and more diverse dataset, including images with subjects wearing glasses, varying ethnicities, and different lighting conditions, can significantly improve generalization. By continuously refining these aspects, the Driver Vigilance Monitoring System has the potential to become an even more indispensable tool in promoting road safety and preventing fatigue-related accidents.

References

[1] National Highway Traffic Safety Administration (NHTSA). (Various Years). Traffic Safety Facts. U.S. Department of Transportation. Available: https://www.nhtsa.gov/ [2] E. Picton and E. K. St Louis, \"Drowsy Driving: A Systematic Review of Behavioral and Physiological Monitoring Technologies,\" Sleep Medicine Reviews, vol. 51, p. 101289, 2020. doi: 10.1016/j.smrv.2020.101289. [3] W. W. Wierwille, W. F. Wreggit, and R. J. Fairbanks, \"Research on Vehicle-Based Driver Fatigue Detection Systems,\" Technical Report, 1994. [4] A. Sahayadhas, K. Sundaraj, and S. Murugappan, \"Detecting Driver Drowsiness Based on EOG, EEG and ECG Signals: A Review,\" IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 38, no. 5, pp. 1056-1069, 2012. doi: 10.1109/TSMCB.2012.2192736. [5] H. Summala and H. Mikkola, \"Fatal Accidents and Their Drivers in Finland,\" Accident Analysis & Prevention, vol. 26, no. 3, pp. 297-304, 1994. doi: 10.1016/0001-4575(94)90011-8. [6] A. K. Ranjani and P. Kavitha, \"The study of driver fatigue monitor algorithm combined PERCLOS and AECS,\" in Proc. 2014 IEEE International Conference on Advanced Communications, Control and Computing Technologies (ICACCCT), Dec. 2014, pp. 1092-1095. doi: 10.1109/ICACCCT.2014.7032128. [7] A. Emin, F. G. Sancak, A. Bayar, and F. Cakar, \"Deep Learning-Based Driver Assistance System,\" ELECTRICA, vol. 23, no. 3, pp. 607-618, 2023. Available: https://electrica.ieu.edu.tr/en/pub/2023_v23_n3_a17/ [8] C. Yang, Z. Yang, W. Li, and J. See, \"An Intelligent Driving Assistance System Based on Lightweight Deep Learning Models,\" IEEE Access, vol. 10, pp. 107062-107071, 2022. doi: 10.1109/ACCESS.2022.3211516.

Copyright

Copyright © 2025 Kondamudi Manoj Kumar, K.S.S. Soujanya Kumari. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.