Discussion on Machine Learning and Deep Learning based Makeup Considered Eye Status Recognition for Driver Drowsiness

Abstract

Driver drowsiness is a primary causes of traffic accidents, may bring great economic and physical harm not only to divers but also to traffic accident victim(s) and their families. Several systems have been developed for detecting the drivers drowsiness. However, few technique consider the case of people who madeup heavily. This paper aims to obtain an effective eye-status recognition method which considers the heavy makeup, by comparing the machine learning and deep learning method. In term of machine learning method, this paper uses histogram of oriented gradients to extract the feature and uses support vector machine to classify the eye-status for judging the drowsiness. About the deep learning method, this paper uses conventional AlexNet and GoogLeNet to classify the eye status. This paper also creates a dataset which considering the makeup and no-makeup, left-eye and right-eye, close-eye and open-eye for experimentation. The experimental results show that GoogLeNet achieves non-mis-recognition in classifying the eye-status under close and open. Machine learning achieves a better recognition rate in judging the left or right eyes than deep learning. The results give us a hint to design a higher accuracy eye status recognition for detecting the driver drowsiness.