Driver Drowsiness detection using Convolutional Neural Network

Driver drowsiness and tiredness are critical factors contributing to road accidents worldwide. Several accidents happen every year because of drowsy driving. This research paper introduces a pioneering approach to tackle driver drowsiness using convolutional neural networks (CNNs) based on the inception V3 architecture in automotive safety systems. The system focuses on monitoring the driver's facial cues, particularly eye movements, to identify instances of prolonged eye closure as indicators of potential drowsiness. The model is trained on a dataset comprising 84,898 images of open and closed eyes from 37 individuals. Training, validation, and test accuracies and losses, along with model architecture details, parameters, and performance metrics, is extensively analyzed and reported. The proposed system aims to enhance driver safety by issuing alerts when the driver exhibits prolonged eye closure, surpassing a predefined threshold. This study outlines the challenges, details the CNN-based model's design, dataset considerations, and showcases its robustness through real-world driving scenario experiments. Emphasizing the efficiency of real-time deployment, this study discusses optimizations and suggests future enhancements. Keywords—convolutional neural network (CNN), transfer learning, deep learning, driver state monitoring(DSM), InceptionV3