Driver fatigue monitoring system

Abstract

Statistics have shown that driver fatigue is one of the leading causes of traffic accidents. Over the past few years, a lot of research effort has been put into designing systems that monitor both driver and driving performance, issuing warnings at the first indications of fatigue to help reduce the number of road fatalities. In this dissertation, a non-intrusive computer vision approach which exploits the driver's facial expression is considered and implemented. A video camera located on the vehicle's dashboard is used to capture the driver's facial region. The developed method employs Viola-Jones classifiers to detect the driver's face, eyes and mouth after performing histogram equalization and median filtering on the original image. Correlation coefficient template matching is then applied on the identified search regions to determine the most likely state of each facial feature in every frame, after which the average time interval between eye closures, the PERcentage eye CLOSure over time (PERCLOS) and the degree of mouth opening are measured for overlapping 20-second fatigue windows. A Support Vector Machine (SVM) is finally integrated within the system to classify the facial appearance as either fatigued or otherwise. Using this simple and cheap implementation, very promising results were obtained, whereby the overall system achieved an accuracy of 95.2%, outperforming other developed systems employing expensive hardware to reach the same objective.