Real-Time Yawn Extraction for Driver’s Drowsiness Detection

Abstract

Driver’s drowsiness is one of the major causes of increase in the number of road accidents. Therefore, design and implementation of a real-time driver’s drowsiness detection system are considered as a crucial component of the Advanced Driver Assistance System (ADAS). Along with other physiological parameters, yawn is often considered as one of the most important indicators of drowsiness in human. Thus, in this work we propose an efficient and nonintrusive system for yawn detection targeted toward real-time drowsiness detection. The proposed algorithmic pipeline consists of a face and facial landmark detector for face detection and landmark localization, a scheme for extracting feature named mouth aspect ratio (MAR) characterizing the state of the mouth (open/close) in each frame, and a classifier to classify the state of the mouth in a temporal window of some fixed number of frames. The performance of the proposed approach has been validated on a manually annotated dataset extracted from the widely used yawn detection dataset called YawDD. The proposed approach has achieved an accuracy of 99.25% along with F1 score of 98.00% and runs at 30 frames per second (FPS).