Analysis of Drowsiness Detection based on Images Using Convolutional Neural Network

Abstract

Drowsiness detection is crucial in maintaining the safety and alertness of individuals, especially in high-risk situations such as driving or operating heavy machinery. This research aims to develop a drowsiness detection system based on facial images using Convolutional Neural Network (CNN) with a focus on the AlexNet method and its comparison with ResNet. In this study, facial image data was collected from various conditions of drowsiness and normal conditions. Image preprocessing was performed to standardize the size and ensure consistent image quality. AlexNet and ResNet were implemented and trained using the image dataset to identify distinctive patterns that differentiate drowsy faces from faces in a normal state. The results of the experiments showed that the use of AlexNet and ResNet methods effectively detects drowsiness in facial images with high accuracy. However, there are performance differences between the two methods. ResNet demonstrated superior performance in certain conditions, while AlexNet showed advantages in other cases. This research contributes to the development of facial image-based drowsiness detection technology applicable in various fields, including smart vehicles and security systems. The comparison results between AlexNet and ResNet also provide valuable insights for selecting the most suitable CNN method for drowsiness detection applications based on facial images.