Research On Infrared Video For Driver Fatigue Detection

In recent years, with the rapid development of transportationindustry and growing car ownership, the number of traffic accident is risingand driver fatigue is one of the important reasons. It is the "invisible killer" intraffic safety area of countries which can not be ignored, so the research onhow to detect and avoid driver fatigue has become the hot spot. Amongnumerous fatigue detection methods, the method based on machine vision hasgood operability and application prospect.Because night is the multiple stages of driver fatigue, this paper presentsan improved driver fatigue detection methods by using a special infrared lightto get the driver’s facial image．The algorithm is divided into four parts: facedetection, face tracking, eye detection and driver fatigue recognition.The main contents are as follows：Face detection is the preliminary work in the driver fatigue detection.This paper presents fast two-dimensional entropy segmentation and regionalconnectivity to locate face, first should segment face from the background,and then accurately locate faces based on the advantage that the face have alarge proportion in image. This method not only has high positioning accuracy,but also has great operating rate to meet the real-time requirements of system.Meanwhile, driver always have small head movement angle in the process ofnormal driving, through the characteristic, this paper adopts an improvedMean-shift algorithm for face tracking to further improve the detection rate ofthe system.Eye detection is an important step in the driver fatigue detection．This paperpresents a cost sensitive support vector machine (SVM) algorithm to detect humaneye. With embedding reject cost and misclassification cost into SVM, thismethod not only inherits the quality of dealing with small sample problem, and also considers the cost factor which traditional classifier do not have. Ithas lower mistake examining rate and improves the reliability of classifier.On the basis of PERCLOS theory, this paper detects human eye closed degree bycalculating eye areas and statistics the proportion that the eye closed frame numberaccounts for the total frame number to monitor driver fatigue in real time．For theinstability of dim light in night, this method uses infrared light as a resource whichhuman eyes are not sensitive．This algorithm is not only simple and effective, but alsohas better tolerance and robustness．...