SRM INSTITUTE OF SCIENCE AND TECHNOLOGY

Ramapuram, Chennai – 600 089

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

18CSP109L- MAJOR PROJECT

Enhancing Driver Safety Through Transfer Learning Models for Detecting
Drowsiness Based on Eye Movements

BATCH NUMBER: 05
Team Members Supervisor
N SIVA SAI DANUSH [RA2111003020129] NAME: Mrs.S.Sajini, AP/CSE
MEDURI NIKHESH [RA2111003020162]
Y V SURYA PRAKASH [RA2111003020173]

Date : 25/01/2025 1
 Agenda

• Abstract
• Literature survey
• Scope and Motivation
• Objectives
• Novel idea
• Block diagram of proposed system
• Tools to be used
• Applications
• References

2
 ABSTRACT
• Driver fatigue causes many road accidents, and traditional detection methods like
EEG and costly sensors are not practical for daily use.
• The project uses Haar Cascade classifiers to track eye movements and identify
drowsiness through the Eye Aspect Ratio (EAR).
• Real-time video is used to analyze facial and eye landmarks, calculate EAR, and
detect drowsiness using machine learning models.
• Alerts are generated immediately when drowsiness is detected to prevent
accidents. The system is designed to be affordable, scalable, and easy to use with
regular cameras, improving road safety.
• It can also be integrated into existing driver-assistance systems for wider
applications. Future versions could support more advanced hardware for higher
accuracy.

3
 LITERATURE SURVEY
S.No Title Authors Year AI/ML Technique Description Advantages Limitations Key Findings

Real-Time Drowsiness Transfer Learning, Multi-modal driver Computational Enhanced safety

1 Chen et al. 2024 94% accuracy
Detection CNN state analysis complexity prediction

Multimodal Drowsiness Multi-sensor Integrated driver Comprehensive High data Robust drowsiness
2 Rodriguez et al. 2023
Detection Framework Transfer Learning behavior analysis monitoring complexity identification

Advanced driver
Deep Learning for Driver State Transformer High computational Precise fatigue
3 Thompson & Lee 2024 behavioral Adaptive learning
Recognition Transfer Learning requirements assessment
prediction

Explainable AI for Driver SHAP, Interpretable Transparent Model Explanation Improved safety
4 Garcia et al. 2023
Drowsiness Models drowsiness detection interpretability complexity insights

4
 SCOPE AND MOTIVATION

• This project aims to create a real-time system to monitor driver safety by detecting
drowsiness using Haar Cascade classifiers and transfer learning models for more accurate
and faster fatigue classification.
• The solution is cost-effective as it works with regular webcams or mobile phone cameras,
eliminating the need for specialized equipment, which keeps costs low and makes the
system easily accessible for everyday use, even for individuals or businesses on a budget.
• Driver fatigue is a leading cause of road accidents worldwide. By monitoring fatigue in a
non-invasive, affordable way, this system can help reduce accidents and save lives..
• Software-based machine learning models make it easier to implement in regular vehicles
and integrate with advanced safety systems, improving road safety in a scalable way.
This approach reduces reliance on costly hardware, making it a viable solution for both
existing and new vehicles..

5
 OBJECTIVES

• Implement a real-time drowsiness detection system using Haar Cascade

classifiers for facial and eye detection.
• Leverage transfer learning models to reduce the training time and data
requirements.
• Use eye aspect ratio (EAR) for detecting drowsiness through eye closure
patterns.
• Provide real-time alerts (visual or auditory) when drowsiness is detected.
Build a low-cost, easy-to-implement solution that works with existing
webcams or mobile phone cameras.

6
 NOVEL IDEA

• Objective:
• Enhance the system's performance under extreme conditions like low light,
glare, and obstructions (e.g., glasses, masks).
• Challenges Addressed:
• Reduced accuracy in low-light environments.
• Glare from glasses or reflective surfaces.
• Partial face or eye obstructions caused by masks or eyewear.
• Proposed Solutions:
• Robust Preprocessing Techniques:
• Adaptive contrast adjustment for low-light enhancement.
• Glare reduction using semantic segmentation and inpainting techniques.
• Focus on upper facial regions when parts of the face are obstructed.

7
 • Lightweight Models:
• Utilize optimized architectures like MobileNet or YOLO Nano for faster
real-time detection on resource-limited devices.
• Benefits:
• Improved accuracy across diverse and extreme scenarios.
• Enhanced system reliability for real-world applications.
• Broader applicability for smart vehicles and healthcare monitoring.

8
BLOCK DIAGRAM OF THE PROPOSED SYSTEM

9
 Tools to be used

• Programming Languages: Python: For implementation.

• Libraries: OpenCV: For Haar Cascade classifiers and image processing.
• TensorFlow/Keras: For transfer learning models. NumPy: For numerical
operations.
• dlib: For accurate face and eye detection.
• Hardware: Webcam: To capture video for detecting eye movements .

10
 APPLICATIONS
• Road Safety: Real-time monitoring of drivers’ alertness to prevent
accidents.
• Fleet Management: Monitor commercial vehicle drivers to ensure their
safety during long-distance driving.
• Driver Insurance: Used by insurance companies to assess risk and reward
safe drivers with discounts.
• Smart Vehicles: Integration with ADAS (Advanced Driver Assistance
Systems) for safer autonomous driving.
• Health Monitoring: The technology could be adapted for use in healthcare,
monitoring patients with sleep disorders or fatigue.

11
 References
[1] H. A. Madni et al., "Novel Transfer Learning Approach for Driver
Drowsiness Detection Using Eye Movement Behavior," IEEE Access, vol.
12, pp. 64765-64778, May 2024.
[2] A. Chen et al., "Real-Time Drowsiness Detection," IEEE Transactions
on Intelligent Transportation Systems, vol. 25, no. 3, pp. 456-472, Mar.
2024.
[3] S. Kumar and R. Singh, "Machine Learning for Driver Fatigue
Monitoring," Transportation Research Part C, vol. 152, pp. 112-129, Feb.
2024.
[4] L. Wang et al., "Edge AI in Driver Safety Systems," IEEE Journal of
Selected Topics in Signal Processing, vol. 18, no. 2, pp. 256-271, Feb.
2024.
[5] M. Rodriguez et al., "Multimodal Drowsiness Detection Framework,"
ACM Transactions on Computing for Healthcare, vol. 5, no. 1, pp. 78-95,
Jan. 2024. 12