Research Of Driver Fatigue Detection Based On Machine Vision And Learning

Driver fatigue is a major cause of traffic accidents today,Thus driver fatiguedetection has become a hot topic of current research.The fatigue detection method basedon machine vision and learning has overcome the shortcoming of poor real time、bettercontact and poor accuracy of traditional methods as a new fatigue detection method.We have analyzed some exsiting fatigue detection methods base on machine visionand proposed an improved fatigue detection system. The main algorithms are：facedetection、eye detection and tracting、the calculation of eye features、mouth positioningand tracting、the calculation of mouth and fatigue recognition based on logistic regressionmodel.The main content are as follows.1.Face location.Fisrt of all, To capture images of the driver’s head through thecamera and use homomorphic filtering method of image preprocessing.On the basis ofAnalyzing several face detection methods,we propose an improved face detectionmethod based on color gaussian model and based on haar-like features face detectionmethod.Experiment results have showed the improved face detection method based oncolor Gaussian model is better than the face detection method based on haar-likefeatures.2.Eye location、eye tracking and the calculation of eye features.On the basis of facelocation,we propose an eye location method based on eye strengthen chart by analyzingtraditional eyes location mehods, use the tracking method based on position offset andcalculate the PERCLOS features、blink frequency and blink duration in specified frames.3.Mouth location、mouth tracking and the calculation of mouth feature. We presentthe mouth location methods based on mouth strengthen chart and the algorithm ofestimating inner lip opening distance based on the classification of mouth states torecognize the opening state of mouth and compute the percentage of mouth closing andmouth open-close frequency.4.Driver fatigue driving recognition based on logistic regression model.By training the samples which are the front fatigue-related features composed of PERCLOS、blinkfrequency、blink duration、the percentage of mouth closing and mouth open-closefrequency to build a two category(tired、tiredless) classification which can recognizewhether the driver is tired or not.Besides,to build another two category(tired、sleep)classification which can recognize whether the driver is tired or sleep.This driver fatiguerecognition system is more accurate than traditional method based on machinelearning,which divides fatigue states into three category to avoid yielding road safetyaccidents by taking different measures.