Driver's Drowsiness Detection System (Based on Eye State)

Abstract

There are numerous potential applications for classifying eye conditions, such as detecting tiredness and evaluating psychological states. Due to its importance, many studies have already utilized standard neural network algorithms, achieving promising results. Convolutional neural networks (CNNs) are particularly effective in real-time applications, offering both high accuracy and speed. Early detection of drowsiness is crucial, as it greatly increases the likelihood of preventing accidents. By leveraging artificial intelligence (AI), drowsiness detection can be automated, enabling the evaluation of more cases in less time and at a lower cost. In this system, modern deep learning (DL) and digital image processing (DIP) techniques were employed, specifically using a CNN model.The CNN model used in our project is MobileNetV1 for eye state classification. The MRL Eye dataset was used to train the model. MRL eye dataset consists of images of 37 different individuals (33 men and 4 women). MRL dataset consists of 47,173 eye images from which we have selected 20,940 eye images to train our model. Additionally,we collected 1452 more eye image data of about 30 different subjects. The dataset was split into training sets and testing sets in the ratio of 80:20. Then this dataset was trained using modified MobileNetV1. After training for 30 epochs the training accuracy was about 0.9967 and validation accuracy was 0.8948. This report explains how we created a drive�s drowsiness detection system that predicts the state of a driver�s eyes to further determine the driver�s drowsy state and alerts the driver before any severe threats to road safety.