In-Vehicle Monitoring for Passengers' Safety

Abstract

Driving drowsiness detection through videos/images is one of the most important issues for driver safety in today's world. Because of the great advancements in technology in the last few decades, deep learning techniques applied to computer vision applications such as sleep detection have shown promising results. Drowsiness is characterised by closed eyes, yawning, and micro-sleeps. Moreover, one of the biggest tragedies in the news lately, is toddlers or pets dying from heat built up in cars. In this work, a real-time deep learning algorithm is designed to monitor driver drowsiness, driver distraction, as well as an alert system for forgetting children and pets, and a seat belt usage system. The approach taken was to recognise and localise the face, eyes, and mouth, using the Dlib library, Histogram of Oriented Gradients, and a facial landmark predictor. The eye aspect ratio and the mouth aspect ratio are then calculated and evaluated for yawning detection and micro-sleep detection. The information on the driver's state was saved using a Firebase real-time database. This information is used by the children and pets detection algorithm, which sends an automatic email to the driver if a child or pet is discovered in the backseat when the driver is not in the car. When a driver uses a cell phone, eats, or drinks while driving, this is considered as a distraction. Canny edge detection is used to monitor the seat belt. Furthermore, the proposed method was subjected to several rounds of testing, that proved its viability and reliability.