Research On Train Driver’s Fatigue Detection Based On PERCLOS

Driving Safety of Train grasp the lifeline of railway transportation and driving fatigue for the train drivers is the most important factor that affect driving safety of train. Nowadays, the question that how to detect driving fatigue for train driver real-time and warn timely, in order to reduce the occurrence of train traffic accidents effectively, has become a hotspot. In this paper, machine vision is used to research a non-contact, accurate, real-time train driver’s driving fatigue detection method, which calculates PERCLOS eye fatigue characteristic value by way of image processing, and then judges the train driver’s fatigue. The research is divided into the following three parts:Firstly, this paper researches a rough-to-fine, fast and accurate method on face detection and location method. In the first place, the method needs to establish the Gaussian model and to segment the skin color area in face image, in order to narrow the possible scope of face and improve the face detection speed. And then, the face classifier is trained with AdaBoost algorithm that is most accurate in face detection. At last, the face classifier is used to locate face accurately in skin color area which has get rid of lots of pixels in background.Secondly, a combination of eye classifier and dynamic human eye template matching method for eye detection and location is proposed. This method takes advantage of continuous eye position and morphology among consecutive image frames. In the first place, the proposed method uses classifier of human eye to locate eye area in the face region accurately and segments sub-image of human eye as a template, and then the template matching method is used to locate the eyes from face area in a number of adjacent frames of image. In order to improve the face detection speed, Kalman filtering method is make use of to predict the position of human eye, the eye position prediction is compared with the next frame skin region to integrate the face and eye detection process. This method greatly reduces the face detection time and achieves a fast and accurate location of face and eye.Thirdly, an eye state identification method is proposed which combines of eye aspect ratio and black and white pixels ratio. The proposed method can identify the eye state fast and accurately in the image according to the differences of eye shape and black and white pixels composition in different opened and closed eye states. The PERCLOS method is selected as the driving fatigue detection method in this paper, the train driver’s eye PERCLOS characteristic value is get in continuous image frames to judge the driving fatigue state of train driver according to P80standard of PERCLOS method.At last, a train driver fatigue detection system is established to verify the effectiveness, accuracy, real-time and robustness of the driver fatigue detection method researched in this paper.