Deep Learning for Real-Time Drowsiness Onset Detection

Drowsiness detection is a pivotal element in ensuring the well-being of both drivers' and passengers safety, as it helps prevent accidents caused by tired or drowsy individuals behind the wheel. This research introduces an advanced hybrid model to effectively and swiftly identify drowsiness in real-time. The process commences with data pre-processing, involving the capture and examination of facial images of the driver for analysis. These images go through different pre-processing stages, including resizing, converting to grayscale, and reducing noise through a blend of filters, such as Weiner and Median filters. Our model similar to ResNet, is meticulously fine-tuned and optimized using sophisticated techniques, including Aquila optimization, to enhance its performance. The primary objective of this model is to ascertain whether the driver is in a drowsy or alert state. The model's performance is rigorously assessed and benchmarked against previous research in the field. The attained outcomes by our proposed drowsiness detection model are truly impressive, boasting an F1 score of 98.98%, an accuracy rate of 99.2%, a sensitivity rate of 98.9%, a specificity rate of 99.1%, and a precision rate of 99%. These exceptional metrics underscore the model's effectiveness in accurately identifying drowsiness, underscoring its importance in promoting road safety.