Driver Fatigue Detection Based On Eye Condition With Deep Learning

Traffic accidents are a potential risk that endanger people's health and life. Considering the increasing number of vehicles in recent years, drivers should be more careful in traffic in order to prevent traffic accidents. Most of the accidents in the world are caused by the driver. In particular, the drivers being tired while on the road and carelessness due to fatigue increase the risk of accidents. Today, with the development of digital image technology, a system that detects fatigue driving and warns the driver during the journey has been considered. With Yolov3, one of the deep learning algorithms, the driver's blink rate and the driver's eye closure time are determined and the driver's fatigue degree is determined. In order to detect the blink status of the driver, the Yolov3 algorithm was trained via Python language with a dataset of 3640 images consisting of two classes called open eyes and sleepy eyes. In the Yolov3 algorithm, which was trained to detect and classify open eyes and drowsy eyes classes, 95.4% precision value, 97.9% recall value and 98.2% mAP value were obtained. The experimental results obtained show that our model correctly detected 224 of 231 drowsy eyes images and 251 of 254 open eyes images. The developed system can be integrated into embedded systems and used in traffic to ensure the safety of drivers