Development of driver safety system through the application of deep learning techniques for drowsiness detection

Abstract

This project presents a driver drowsiness detection system based on facial image analysis using deep learning. The goal is to automatically identify whether a driver is alert or drowsy by analyzing behavioral signs such as eye closure, yawning, and head position.We used the UTA-RLDD dataset, which contains real facial images of drivers in both drowsy and non-drowsy states. After preprocessing the images, we applied transfer learning with four pre-trained models : ResNet50, MobileNet, DenseNet201, and EfficientNetV2.The best results were achieved using EfficientNetV2, which reached an accuracy of 99.98%. Our system was evaluated using standard performance metrics such as precision, recall, F1-score, and AUC. This work shows that combining computer vision with deep learning can provide an accu- rate and practical solution for detecting driver fatigue, helping reduce the risk of accidents.