A Study Of Driver Fatigue Detection Based On Machine Vision

Driver fatigue has become one of the significant factor in transportation. Driver fatigue detection has become a hot topic of current research. Fatigue detection based on machine vision, by the advantages of real time and non-contact, becomes a main method of the fatigue detection.Based on the analysis of existed research methods, this paper designs a set of improvement fatigue detection system, which includes four main algorithms:face location, facial features region location and tracking, eyes location, eye state. identification and fatigue detection. Main research methods are as follows:(1) Face location. Reads an image from the infrared video stream image, uses the iterative thresholding algorithm binarize the image, uses the method of region mark to remove inface region. Uses an improved method of integral projection to locate the face region, simultaneously according to the human face proportion characteristic, proposes one kind of the second face localization algorithm. This method not only can make the face location faster, but also can solve the weaknesses of tradition algorithms locate the face region failure affected by shoulder and neck.(2) Facial features region location and tracking. Before locating human eyes, should locate the facial features region. This paper uses the method of gradient maximum value to obtain facial features region, the positioning accuracy using this method to locate facial features region is high, but high complexity. In order to improve the overall system speed, this paper uses a method based on displacement to track facial features. This algorithm is low complication and can solve the shortcomings of using gradient algorithm to locate the facial features slowly.(3) Human eyes location. According to the characteristics of the pupil's gray intensely changed in the infrared images, uses Robert edge detection in which way by calculating the maximum gray level to locate the human eyes. This method is easy and effect. The algorithm of facial features region tracking can remove the eyebrow area reducing the interference factor which not only can make face location faster, but also can improve the accuracy of eye position to meet the system in real time and to meet the requirements of the system accuracy.(4) Eye state identification and fatigue detection. Based on detailed analysis of the characteristics of the eye projection diagram, this paper puts forward a second eye recognition algorithm based on the proportion of projection. This method not only can retain the projection merit of fast and simple, but also can solve weaknesses of the Low recognition rate of tradition projection. After eye state identification, by PERCLOS (percentage of eyelid closure over the pupil over time) and blink frequency combination method, the fatigue can be detected.