Systematic development of real-time driver drowsiness detection system using deep learning

Abstract

Advancements in globalization have significantly seen a rise in road travel. This has also led to increased car accidents and fatalities, which become a global cause of concern. Driver's behavior, including drowsiness, contributes to many of the road deaths. The main objective of this study is to develop a system to diminish mishaps caused by the driver's drowsiness. Recently deep convolutional neural networks have been used in multiple applications, including identifying and anticipate driver drowsiness. However, limited studies investigated the systematic optimization of convolutional neural networks (CNNs) hyperparameters, which could lead to better anticipation of driver drowsiness. To bridge this gap, a holistic approach based on the deep learning method is proposed in this paper to anticipate the drivers' drowsiness and provide an alerting mechanism to prevent drowsiness related accidents. To ensure optimal performance achievement by the system, a database of real-time images preprocessed via Haar cascade's classifiers is used to systematically optimize the CNN model's hyperparameters. Different metrics, including accuracy, precision, recall, F1-score, and confusion matrix, are used to evaluate the performance of the model. The training evaluation results of the optimal model achieved an accuracy of 99.87%, while the testing results accurately classify the drowsy driver with 97.98%.