DRIVER DROWSINESS MONITORING USING CNN

Abstract

Many safety connected driving supporter schemes decreased the danger of four-wheeler accidents,
and investigations depicted weariness to be a major reason of four wheeler accidents. A car organization
announced an idea that whole deadly accidents (17%) would be attributed to weary drivers. Many revisions
showed by Volkswagen AG specify that 5-25% of all accidents are produced by the sleeping of driver. The lack
of concentration damage steering actions and decrease response period, and revisions illustrated that sleepiness
raises threat of crashes demand for a dependable intelligent driver sleepiness sensing system. The aim is to
create an intelligent processing scheme to avoid road accidents. This can be done by period of time monitoring
the drowsiness and warning driver of inattention to prevent accidents. Based on the literature survey, the driver's
drowsiness can be detected based on three factors such as physiological, behavioral, and vehicle-based
measurements. But these approaches pose some disadvantages in certain real-time scenarios. So, we aim to
apply Deep Learning algorithms to this problem statement. Our methodology is to use a Convolutional Neural
Network (CNN). CNN offers a computerized and effective way to categorize the driver as drowsy or nondrowsy
correctly The advancement in computer vision has assisted drivers in the form of automatic self-driving
cars etc. The misadventure is caused by driver's fatigue and drowsiness about 20%. It poses a serious problem
for which several approaches were proposed. However, they are not suitable for real-time processing. The major
challenges faced by these methods are robustness to handle variation in human face and lightning conditions. We
aim to implement an intelligent processing system that can reduce road accidents drastically. This approach
enables us to identify driver's face characteristics like eye closure percentage, eye-mouth aspect ratios, blink rate,
yawning, head movement, etc. In this system, the driver is continuously monitored by using a webcam. The
driver's face and the eye are detected using haar cascade classifiers. Eye images are extracted and fed to Custom
designed Convolutional Neural Network for classifying whether both left and right eye are closed. Based on the
classification, the eye closure score is calculated. If the driver is found to be drowsy, an alarm will be triggered.