A Real-time Fatigue Detection System using Multi-Task Cascaded CNN Model

Drowsiness is a leading cause of automobile collisions, with significant consequences for road safety. Driver fatigue is a significant factor in the large number of crashes. The World Health Organization (WHO) estimates that 1.22 million traffic accidents occur every day. If drowsy drivers are given adequate notice, several crashes can be avoided. Support Vector Machines are used in a variety of drowsiness detection methods to monitor drivers' fatigue levels when driving and warn them if they are not focusing on the road (SVM). Facial movements can then be used to retrieve more accurate characteristics including percentage of eyelid closure over the pupil over time (PERCLOS), yawning, and head nodding to infer the degree of drowsiness. This paper proposes a new approach to analyse and intimate driver drowsiness using Deep Learning techniques based on Convolutional Neural Networks (CNN). Results indicate that the proposed MTCNN architecture is capable of achieving an accuracy rate of 97.5% .