Enhancing Driver Safety with MobileNetV2-Based Transfer Learning for Drowsiness Detection

Driving requires consistent concentration, and distracted driving can lead to severe accidents. Every year, the number of traffic accidents globally continues to rise, often due to driver fatigue and drowsiness. To address this critical issue, numerous researchers have gathered real-time driver drowsiness data and applied various deep learning algorithms and feature combinations to improve the efficacy of detection systems in real-time. This paper aims to create a machine learning model capable of detecting whether a driver’s eyes are open or closed and sounding an alert or warning if it detects signs of drowsiness. To achieve this, we utilized transfer learning, a method that leverages pre-trained models on large datasets to enhance the performance of our specific task. Transfer learning can significantly reduce training time and improve accuracy by building on existing knowledge from related tasks. The model processes real-time video feeds from a camera mounted in the vehicle, analyzing the driver’s eye state continuously. If the model detects that the driver’s eyes remain closed for a certain period, it will trigger an alert, thereby preventing potential accidents caused by drowsiness. This research involves collecting a substantial dataset of eye images labeled as open or closed, training the model using advanced convolutional neural networks (CNNs), and fine-tuning it for optimal performance. The proposed model achieved an accuracy of 98.5%. The approach encounters difficulties including inconsistent lighting conditions, dependence on camera quality, and limitations in computational capacity. By implementing this system, we aim to reduce the incidence of drowsiness-related collisions, thereby enhancing overall traffic safety. This approach not only promises to safeguard drivers and passengers but also contributes to the broader goal of developing smarter and more responsive automotive safety technologies.