Drowsiness Detection in Drivers with Explainable Artificial Intelligence

Drowsiness driving, which can be defined as the desire to rest-induced condition of sleepiness, can result in symptoms that significantly affect how well driving task is performed, such as sluggish response times, brief loss of awareness, or microsleeps (blinks lasting longer than 500[Formula: see text]ms), to mention a few. In this sense, drowsy driving is a major safety concern on roads worldwide, as it can lead to serious injuries and fatalities. In this paper, we propose the development of a model for detecting drowsiness in drivers from images that focus on the driver’s face. The model is created through the use of Deep Learning, more specifically, using Convolutional Neural Networks. For this purpose, RLDD Dataset, which consists of around 30[Formula: see text]h of RGB categorized videos of 60 drivers, was used. In addition, the most important contribution in this paper is that the decisions of the model generated are interpreted and explained through the application of different Deep Learning Explainable techniques. As a result, we propose a drowsiness detection model with high accuracy, enhanced by the integration of multiple visual explanation techniques that deliver promising results.