Abstract
Automotive industry experiences multiple injuries in our everyday life. The increasing road accident rate is due to driver drowsiness, such as fatigue and insomnia. This research is intended primarily to diagnose exhaustion and drowsiness, utilizing the deep learning models like convolution neural network. A research project has been initiated in the Kingdom of Saudi Arabia to address the problem of identifying the driver's drowsiness state. In this article, we present a real-time driver disturbance monitoring method using Convolutional Neural Network (CNN). Detection is done using state-of-the-art CNN models such as InceptionV3, VGG16, and ResNet50. We have analyzed that InceptionV3 has 90.70% with 0.6931 loss, the VGG16 has 39.87% with 0.6931 loss, and the ResNet50 has 93.69% accuracy with 0.6931 loss. So, the ResNet50 model has the best accuracy as compared to all other deep learning models. Using sleepy and active human face image collection, we present the output review of the CNN models. We expanded device deployment and performance review of the best-performing CNN model among the evaluated models, utilizing our custom dataset to enhance real-time efficiency. This unique data collection is similar to the driver end scenarios because it includes not just the side view, but also the driver's front view, greatly enhancing the performance. Extending this work into a commodity and its large-scale production will also render a significant contribution to the Kingdom's economy.