Advancing Road Safety: Heuristic and Learning-Based Methods for Robust Driver Drowsiness Detection

Additional Authors

Manoj Adhikari, Department of Computing, Collge of Business And Technology, East Tennessee State University, Johnson City, TN

Abstract

INTRODUCTION: Driver drowsiness is a critical factor contributing to road accidents globally, posing significant risks to road safety. Drowsiness detection methods of physiological and vehicular-based approaches are often costly and intrusive which limits their practicality. This research explores cost-effective, non-intrusive methods for drowsiness detection based on drivers' behavioural approach, leveraging advancements in computer vision and machine learning. METHODS: The study presents two complementary approaches: (1) a heuristic-based method that relies on facial indicators like eye closure, yawning, head pose, and eye gaze, analyzed through OpenCV, Dlib, and MediaPipe. The second approach adopts (2) a learning-based method that utilizes machine learning (ML) and deep learning (DL) techniques. This includes transfer learning on pre-trained CNN models (VGG16 and ResNet50) and developing traditional classifiers (Random Forest, SVM, Logistic Regression, XgBoost) based on facial features extracted via MediaPipe. RESULTS & CONCLUSION: The heuristic-based approach achieved 97% accuracy using MediaPipe, outperforming Dlib’s 89%, with superior performance in diverse lighting conditions and with accessories like glasses. For the learning-based approach, ResNet50 achieved 95% validation accuracy, while Random Forest classifiers attained 97%, demonstrating high precision and recall for drowsy states. The system achieved real-time processing with a latency of ~1.2 seconds. This study highlights the effectiveness of combining heuristic-based and learning-based techniques for robust, real-time drowsiness detection. Future work aims to enhance system robustness under varying lighting conditions, integrate multimodal data, and further reduce latency to improve responsiveness. The findings underscore the potential of advanced driver monitoring technologies in mitigating drowsy driving risks and enhancing road safety.

Start Time

16-4-2025 1:30 PM

End Time

16-4-2025 2:30 PM

Room Number

303

Presentation Type

Oral Presentation

Presentation Subtype

Grad/Comp Orals

Presentation Category

Science, Technology and Engineering

Faculty Mentor

Shehenaz Shaik

This document is currently not available here.

Share

COinS
 
Apr 16th, 1:30 PM Apr 16th, 2:30 PM

Advancing Road Safety: Heuristic and Learning-Based Methods for Robust Driver Drowsiness Detection

303

INTRODUCTION: Driver drowsiness is a critical factor contributing to road accidents globally, posing significant risks to road safety. Drowsiness detection methods of physiological and vehicular-based approaches are often costly and intrusive which limits their practicality. This research explores cost-effective, non-intrusive methods for drowsiness detection based on drivers' behavioural approach, leveraging advancements in computer vision and machine learning. METHODS: The study presents two complementary approaches: (1) a heuristic-based method that relies on facial indicators like eye closure, yawning, head pose, and eye gaze, analyzed through OpenCV, Dlib, and MediaPipe. The second approach adopts (2) a learning-based method that utilizes machine learning (ML) and deep learning (DL) techniques. This includes transfer learning on pre-trained CNN models (VGG16 and ResNet50) and developing traditional classifiers (Random Forest, SVM, Logistic Regression, XgBoost) based on facial features extracted via MediaPipe. RESULTS & CONCLUSION: The heuristic-based approach achieved 97% accuracy using MediaPipe, outperforming Dlib’s 89%, with superior performance in diverse lighting conditions and with accessories like glasses. For the learning-based approach, ResNet50 achieved 95% validation accuracy, while Random Forest classifiers attained 97%, demonstrating high precision and recall for drowsy states. The system achieved real-time processing with a latency of ~1.2 seconds. This study highlights the effectiveness of combining heuristic-based and learning-based techniques for robust, real-time drowsiness detection. Future work aims to enhance system robustness under varying lighting conditions, integrate multimodal data, and further reduce latency to improve responsiveness. The findings underscore the potential of advanced driver monitoring technologies in mitigating drowsy driving risks and enhancing road safety.