Abstract
Drowsiness while driving is a serious issue worldwide, threatening the lives of drivers and other road users. It not only results in property damage but also causes injuries and fatalities. Numerous experimental studies have collected data on driver drowsiness and applied various methods to develop warning systems for drivers when they feel fatigued behind the wheel. However, the accuracy of drowsiness prediction can decline if the studies focus solely on facial features such as the eyes and mouth, neglecting other signs of fatigue. To improve the accuracy and detection time of drowsiness from previous research, I propose an effective and precise method for detecting driver drowsiness using enhanced deep learning networks such as MobileNetV2, InceptionV3, ResNet152V2, NASNetLarge, DenseNet121, Long Short-Term Memory, and Vision Transformer, with five levels based on Katajima's scale. Additionally, the proposed model integrates Mostafa's emotion detection model to reduce computational costs and increase accuracy in detecting driver drowsiness. Experimental results demonstrate that the proposed method can achieve up to 98.94% accuracy in various contexts, including wearing masks, using glasses, low light conditions, and normal conditions.