Predicting Driving Behaviour Using Deep Learning

Abstract

In recent years, the rise in automobiles and drowsiness has been a significant cause of accidents, leading to numerous injuries and even deaths. To combat this issue, computerization has been implemented in various areas, promoting uniformity and enhancing the quality of life for users. However, despite the creation of drowsiness-finding systems over the past decade, these systems still require improvement in terms of efficiency, cost, speed, and accuracy, among other factors. This paper proposes an integrated approach that incorporates various parameters, including the PERCLOS eye and mouth check status, the computation of a new vector called the Facial Aspect Ratio (FAR), as well as EAR and MAR, to detect drowsiness. The system detects uncontrolled eye movements, an open mouth, and other actions such as nodding and hand motions to control drowsiness. Additionally, the system includes styles and textural-based grade patterns to detect sunglasses on the driver's face and locate the driver's face in different directions. The proposed study demonstrated, improved precision and tested on datasets similar to NTHU-DDD, YawDD, and EMOCDS (Eye and Mouth Open Close Data Set). An android application utilizing the device's camera can detect drowsiness by observing the user's eyes and face, which can be helpful while driving, working, or studying. In conclusion, this integrated approach offers promising results for detecting drowsiness and enhancing user safety.