Citation

Gupta ND, Rajoo R, Jacob PJ. Malays. J. Comput. 2023; 8(1): 1375-1387.

Copyright

(Copyright © 2023, Faculty of Computer & Mathematical Sciences, Universiti Teknologi MARA)

DOI

10.24191/mjoc.v8i1.20286

PMID

unavailable

Abstract

Driver drowsiness or fatigue is a significant factor that causes road accidents each year and considerably affects road safety. According to the World Health Organization (WHO), drowsy driving may contribute to approximately 6% of fatal and severe road accidents. To overcome this problem, we present a state-of-the-art, real-time drowsiness detection system, which exploits innovative deep-learning techniques to evaluate facial expressions. Our system analyzes not just the driver's eyes, mouth, and head rotation pose with front angles but also left and right yaw angles up to 90° to ensure the driver's safety. We gathered a dataset from public stock image websites, and manual image captures to develop the system. After processing the dataset, we extracted a wide range of features, which we fed into a deep convolutional neural network (CNN) algorithm. Specifically, we employed three different CNN algorithms which are EfficientDet D0, SSD MobileNet V2, and SSD ResNet50 V1, to classify the driver's drowsiness status using the facial key attributes in real time. Our results show that the SSD ResNet50 V1 model exhibited the highest accuracy and consistency in detecting driver drowsiness, underscoring the potential of our innovative system in promoting road safety. Our future work will focus on fine-tuning the approach to enhance its accuracy and performance.

Language: en