Detecting Abnormal State of Driver Using Eye and Head Based on HOG and Landmarks

Traffic accidents are a problem that seriously threatens human life. The Association for Safe International Road Travel (ASIRT) points out that traffic accidents rank 9^th among the many causes of human death. One of the causes of human death is traffic accidents. The main reason is drowsy driving. A difficult step in detecting a sleepy driver is solving the problem of human faces that are not aligned with the receiving camera. This paper aims to develop several detection techniques to help wake drivers in a drowsy state based on image processing techniques. To overcome this problem, we propose a method of head direction estimation based on the facial normal model to detect drowsiness based on the driver's head gestures. In addition, we also use the eye ratio calculation method to detect whether the driver is falling asleep. The results show that the average accuracy of the algorithm reaches 96%. In particular, the accuracy reaches 97% and 93% for strong and low light, respectively.