Real Time Prevention of Driver Fatigue Using Deep Learning and MediaPipe

Abstract

This paper describes the development of a system for detecting driver drowsiness whose goal is to alert drivers of their sleepy state to prevent traffic accidents. It is essential that drowsiness detection in a driving environment be conducted in a non-intrusive manner and that the driver not be troubled by alerts when they are not sleepy. We make use of the MediaPipe Facemesh framework to extract facial features and the Binary Classification Neural Network to precisely detect drowsy states in our solution to this open problem. The solution that minimize false positives is created to determine whether or not the driver exhibits sleepiness symptoms. The approach extracts numerical features from images using deep learning techniques, which are then added to a fuzzy logic-based system. This system typically achieve 91% accuracy on training data and 92% accuracy on test data. The fuzzy logic-based approach, however, stands out because it doesn't raise erroneous alerts (percentage of correctly identified footage where the driver is not tired). Although the findings are not particularly satisfying, the recommendations offered in this study are promising and may be used as a strong platform for future work.