A CNN-based Deep Learning Framework for Driver’s Drowsiness Detection

Abstract

Accidents are one of the major causes of injuries and deaths worldwide. According to the WHO report, in 2022 an estimated 1.3 million people die from road accidents. Driver fatigue is the primary factor in these traffic accidents. There are a number of studies presented by previous researchers in the context of driver’s drowsiness detection. The majority of earlier strategies relied on image processing systems that used algorithms to identify the yawning, eye closure, and eyebrow of the driver taken from the live video camera. One of the major issues of the previous studies was the delay in detection time and dataset. These studies used physical sensors for monitoring the driver’s behavior causes in delay time of detection. In this article, a deep learning approach is used to provide a continuous strategy for detecting driver’s drowsiness using an efficient dataset. The trained algorithm is employed on the video taken from the live camera to extract the driver's facial landmarks, which are subsequently processed by a trained algorithm to provide results. The dataset used for training the CNN algorithm is consisting of 2904 images taken from various subjects under various driving circumstances. The data is preprocessed by different methods including statistical moments, CNN filters, frequency vector determination and position Incidence vector calculation. After training the algorithm the feature-based cascade classifiers files are used to recognize the face from the real-life scenario using the live camera. The accuracy of the purposed model is 95%, which is the highest of all the purposed models, based on data gathered from different kind of scenarios.