Driver Drowsiness Detection

Abstract

This paper presents a real-time, low-cost
driver drowsiness detection system based on visual
behavioral analysis and machine learning techniques.
The system captures live video using a webcam and
computes key facial features such as eye aspect ratio,
mouth opening ratio, and nose length ratio. An
adaptive thresholding method is employed to detect
signs of drowsiness from these features in real time.
Feature values are stored and analyzed using machine
learning classifiers, including Bayesian, Fisher’s
Linear Discriminant Analysis (FLDA), and Support
Vector Machine (SVM). Experimental results indicate
that FLDA and SVM outperform the Bayesian
classifier, with sensitivities of 0.896 and 0.956
respectively, and perfect specificity (1.0) for both. These
promising results suggest that FLDA and SVM are
suitable for real-time implementation. Future work
includes deploying the system on portable hardware
and conducting real-world testing with drivers to
further validate its effectiveness.