Non-Intrusive Drowsiness Detection for Accident Prevention Using a Novel CNN

Abstract

Drowsiness detection problem is not onlycomplex, but also very important for accident prevention.In this paper, we propose a non-intrusive drowsinessdetection method using the right eye and mouth. Facedetection is performed using HOG + SVM methodand facial features are segmented using 8 landmarksobtained by an ensemble of regression trees andclassified using a novel convolutional neural network thatwe call Dozy-Net. Then, drowsiness detection is carriedout using three behavioral parameters: PERCLOS, blinkfrequency, and yawning duration. Two state-of-the-artand one self-constructed dataset were used to train,test, and compare Dozy-Net’s performance against othersix state-of-the-art convolutional neural networks, beingDozy-Net significantly faster. Drowsiness detectionmodel was tested on a real-life dataset performing75.8% accuracy and an average speed of 21.51 FPS.Compared to other existing models, our proposal has theadvantage of having been tested in conditions similar tothose to be expected in a real environment.