Improved Driver Drowsiness Detection Model Using Relevant Eye Image’s Features

Abstract

Driver drowsiness is one of the critical causes of roadways accidents nowadays. Thus fatigue and drowsiness detection play vital role in preventing the corresponding road accidents. Over the last decade, many image processing based approaches were developed to detect driver’s fatigue and drowsiness status. These approaches mainly focus on the extraction of the driver's face and predict the eye blinking rate from the eye region and the yawning rate from the mouth region. However, these features are not necessarily the best to describe the driver’s fatigue level, because from one side, some drivers may have imbalanced eye blinking rate due to medical issues, and from the other side, some drivers may have high yawning rate while they have fully driving attention. In this paper, an online face monitoring system was installed and a large list of eyes area features was extracted in spatial and frequency domain including two new features which are circularity and black ratio. Four support vectors machine classification models were developed based on combinations of the relevant features. The analysis of these models showed that the highest accuracy (91.3%) was achieved when the wavelet coefficients, texture features, circularity, and black ratio are employed. The results of the proposed approach indicated its promising inline implementation into car cabin to decide the driver's drowsiness status.