Vision-Based Driver Drowsiness Detection Using Convolutional Neural Networks: Performance Evaluation and Comparative Analysis

Abstract

The global concern over road accidents due to driver fatigue high-lights the need for efficient detection technologies to improve road safety. The practical constraints of traditional methods that rely on physiological signals have prompted researchers to investigate the use of convolutional neural net-works (CNNs) for the purpose of efficient detection of drowsiness. This paper describes a detailed approach to identifying driver fatigue using a convolutional neural network (CNN). The dataset used includes annotated photos of drivers' faces and Haar cascade eyes for accurate model construction and evaluation. Many data preprocessing methods, such as scaling, normalization, and data augmentation, are used to enhance the performance and generalizability of model. The model architecture consists of various layers for effective feature extraction and classification. The process of training entails the adjustment of hyperparameters, such as a learning rate scheduler, Adam optimizer, and early stopping mechanism, to enhance model convergence and mitigate the risk of overfitting. Furthermore, a comprehensive review of previous studies on detect-ing driver drowsiness with convolutional neural networks (CNNs) shows vary-ing levels of accuracy, ranging between 80% and 98%. Expanding upon the findings, our research introduces a resilient convolutional neural network (CNN) methodology that exhibits a remarkable accuracy rate of 99%. This out-come underscores the considerable potential for enhancing road safety.