An Intelligent Framework for Driver Fatigue Detection Based on Hybrid Model of Gradient Boost Decision Tree-CNN-LR

Drowsy driving is a major cause of fatalities and serious injuries on the roads. Research into methods of detecting and signaling driver weariness is ongoing. Common practices are grounded in either vehicle dynamics, driver psychology, or physiological research. Other ways are intrusive and can be a distraction for the driver, while others make use of pricey sensors and a large amount of data. As a result, this research presents a less costly real-time Drivers Drowsiness detection system that achieves comparable accuracy. Eye and mouth detection are used to calculate the eye aspect ratio and the mouth aspect ratio, respectively, from a facial image. Using adaptive thresholding methods, the returning values are classified to determine whether or not the person is sleepy. The hough transform is used to identify the eye's position after being preprocessed with a gaussian filter. The model is trained with GBDT-CNN-LR after features are extracted using Ostu segmentation. When compared to two other models, including CNN and LR, the proposed method performs admirably.