A CNN-Based Approach for Driver Drowsiness Detection by Real-Time Eye State Identification

Drowsiness detection is an important task in road safety and other areas that require sustained attention. In this article, approach to detect drowsiness drivers presented, focusing on the eye region, since fatigue one of first symptoms drowsiness. The method used for extraction region Mediapipe, chosen its high accuracy robustness. Three neural networks were analyzed based InceptionV3, VGG16 ResNet50V2, which implement deep learning. database NITYMED, contains videos with different levels three evaluated terms accuracy, precision recall detecting region. results study show all convolutional have particular, Resnet50V2 network achieved highest a rate 99.71% average. For better visualization data, Grad-CAM technique used, we obtain understanding performance algorithms classification process.