An Efficient Vision based Method for Detecting Drowsiness in Real-time

Abstract

The driver feeling drowsy while driving is a common situation which has become a major reason for the cause of accidents on road. A drunk person and a person in drowsy state are in the same condition with blur eyes and have slow thought process. There has been various research carried out to detect the drowsy state. Haar Cascade classifiers method has been used a lot in the previous studies for face detection; however, the problem with haar cascade is that they are highly liable to wrong face detections and need adjustment of parameters to detect MultiScale function. This paper proposes a system that utilizes Python, OpenCV and Dlib face detector which uses HOG and Linear SVM. Dlib is a library written in C++, which is highly suitable for solving complex problems. In this paper, Dlib is used for detecting faces and facial landmarks. The proposed algorithm is a real time algorithm to check the eye status from the video input captured in the real time. It considers the position of the landmarks detected on the face and extract the eye region by using both eyes' facial landmarks. The Eye Aspect Ratio is calculated and depending upon the ratio, the system decides to alarm or not. The proposed system gives better accurate results compared to the Haar cascade.