Machine Learning in Driver Drowsiness Detection - A Focus on HRV, EDA, and Eye Tracking - thesis

Aug 5, 2026

Upload by Finn

Drowsy driving contributes significantly to road traffic accidents, making reliable detection systems essential. This research advances driver drowsiness understanding by analyzing physiological indicators, including heart rate variability (HRV), PERCLOS, blink rate, blink percentage, and electrodermal activity (EDA). Data were collected from 40 participants under controlled nonmonotonous and monotonous driving scenarios, while fatigue was measured twice using the Fatigue Assessment Scale (FAS). Three machine learning models were developed and evaluated using comparisons to multiple advanced architectures.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages91

Reading Time3 h 49 min

AccessFree to download

Introduction

Motivation

Research objectives

Thesis overview

Background

Drowsiness causes and effects

Drowsiness measurement techniques

Overview of the variables to analyze

Percentage of eyelid closure over time (PERCLOS)

Blink percentage (BP)

Machine Learning in Driver Drowsiness Detection: A Focus on HRV, EDA, and Eye Tracking by Jose A. Alguindigue Ruiz A thesis presented to the University of Waterloo in fulfillment of the thesis requirement for the degree of Master of Applied Science in Systems Design Engineering Waterloo, Ontario, Canada, 2023 © Jose A. Alguindigue Ruiz 2023 Author’s Declaration I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required final revisions, as accepted by my examiners. I understand that my thesis may be made electronically available to the public. Abstract Drowsy driving continues to be a significant cause of road traffic accidents, necessitating the development of robust drowsiness detection systems. This research enhancesour understanding of driver drowsiness by analyzing physiological indicators – heart rate variability (HRV), the percentage of eyelid closure over the pupil over time (PERCLOS), blink rate, blink percentage, and electrodermal activity (EDA) signals. Data was collected from 40 participants in a controlled scenario, with half of the group driving in a nonmonotonous scenario and the other half in a monotonous scenario. Participant fatigue was assessed twice using the Fatigue Assessment Scale (FAS) . The research developed three machine learning models: HRV-Based Model, EDABased Model, and Eye-Based Model, achieving accuracy rates of 98.28%, 96.32%, and 90% respectively. These models were trained on the aforementioned physiological data, and their effectiveness was evaluated against a range of advanced machine learning models including GRU, Transformers, Mogrifier LSTM, Momentum LSTM, Difference Target Propagation, and Decoupled Neural Interfaces Using Synthetic Gradients. The HRV-Based Model and EDA-Based Model demonstrated robust performance in classifying driver drowsiness. However, the Eye-Based Model had some difficulty accurately identifying instances of drowsiness, likely due to the imbalanced dataset and underrepresentation of certain fatigue states. The study duration, which was confined to 45 minutes, could have contributed to this imbalance, suggesting that longer data collection periods might yield more balanced datasets. The average fatigue scores obtained from the FAS before and after the experiment showed a relatively consistent level of reported fatigue among participants, highlighting the potential impact of external factors on fatigue levels. By integrating the outcomes of these individual models, each demonstrating strong performance, this research establishes a comprehensive and robust drowsiness detection system. The HRV-Based Model displayed remarkable accuracy, while the EDA-Based Model and the Eye-Based Model contributed valuable insights despite some limitations. The research highlights the necessity of further optimization, including more balanced data collection and investigation of individual and external factors impacting drowsiness. Despite the challenges, this work significantly contributes to the ongoing efforts to improve road safety by laying the foundation for effective real-time drowsiness detection systems and intervention methods. Acknowledgements In the completion of this thesis, I am deeply indebted to many people who have made this journey possible and enriched my experience. I would like to express my profound gratitude to my supervisors, Prof. Siby Samuel and Dr. Apurva Narayan, for their continuous support and guidance. Their insight and expertise have been invaluable, and their encouragement and faith in my abilities have greatly inspired me. I am privileged to have learned from their vast knowledge and to have been mentored by them. My sincere appreciation also goes to my reading committee, Dr. Shi Cao and Prof. Oliver Schneider. Their critical feedback and constructive suggestions have been instrumental in shaping this thesis. I have greatly benefited from their expe

FAQs

Which physiological signals are used to detect driver drowsiness?

The research uses heart rate variability (HRV), PERCLOS, blink rate, blink percentage, and electrodermal activity (EDA) signals.

How was the data collection scenario structured for participants?

Forty participants were tested in a controlled setting, with half driving in a nonmonotonous scenario and the other half in a monotonous scenario.

What machine learning models were developed and how did they perform?

Three models were built: an HRV-Based Model (98.28% accuracy), an EDA-Based Model (96.32% accuracy), and an Eye-Based Model (90% accuracy).

Why was the Eye-Based Model less effective than the others?

Its performance was constrained by dataset imbalance and underrepresentation of certain fatigue states, potentially influenced by the limited 45-minute study duration.

Aug 5, 2026

Upload by Finn