A Visual Fusion Based Deep Learning approach for Real Time Driver Drowsiness Detection

Abstract

Growing  concerns  due  to increased road accidents due to driver drowsiness have paved the way  for the development  of  real time driver sleepiness detection systems.The real world implementa -tion of  these systems can play a major role in reducing the accidental rate due to driver sleepin -ess. This paper suggests a  fusion    based  approach  for  identifying  the  driver’s  sleepiness  in  real-time.The model uses driver’s  behavioral  features  like  eye  state  and yawning as a metric for sleepiness identification. Mouth Aspect  ratio  is  used  for  yawn  detection  and  Convolution  Neural Network for eye state classification. Dlib’s HOG based detection is used for facial feature  detection in  this paper. Our proposed model shows a  high  accuracy  rate  and  can detect  if  the  driver  is drowsy  or  not. It  performs  well  in  various lighting conditions  and  also if the driver wears  spectacles. Training   and   testing  accuracy  of  the proposed  CNN  model  is 98.75% and 97.65% respectively. Moreover, it is found that  our  proposed methodology performs  well in real time.