The Research On Fatigue Driving Algorithm For Human Eye Feature Detection

The rapid economic development has led to the rapid development of the transportation industry.The rising number of vehicles has led to a sharp increase in the incidence of vicious road traffic accidents in the country and an upward trend.The losses to society and the family cannot be ignored.Post-mortem investigations revealed that accidents caused by fatigue driving accounted for a high percentage of total traffic accidents.Therefore,the research on driver fatigue detection has great value and practical significance.With the continuous upgrading of computer hardware and software and the rise of neural network technology,fatigue-driving research based on computer vision has become a hot spot in fatigue detection because of its non-contact characteristics,and among all the external features of the driver,people The state of the eye can reflect the fatigue condition very well,and its effectiveness has been proved.Therefore,based on computer vision technology,this paper extracts the eye characteristics of the driver and judges the fatigue according to the state of the eye.In order to develop a set of fatigue driving recognition algorithms that are easy to use and have both real-time and accuracy,the main contents of this paper are as follows:1.Traffic accidents often occur between milliseconds,and the system has high requirements for real-time performance.In order to effectively improve the detection speed,this paper adopts the method of face detection before the human eye is positioned for the detected image.By obtaining the face area to a certain extent,the search area of the eye detection is reduced,thereby effectively shortening the detection time.On the basis of a large number of reading and researching common techniques,this paper finally selects the Haar-Like feature-based Adaboost algorithm to detect the human face region.This technology can balance the accuracy and real-time.2.In order to effectively reduce the range of the image to be detected before the human eye is positioned,and improve the detection rate and accuracy,after detecting the face,combined with the facial a priori knowledge of the "three courts and five eyes" and the horizontal integral projection method The human eye performs rough positioning,effectively eliminates interference information and reduces the detection range of the target area,and improves the detection rate and detection speed.Then combined with the advantages and disadvantages of Sobel operator edge detection and Adaboost algorithm in human eye localization,the two methods are combined to detect human eyes,and the detection method has high accuracy.3.The degree of human eye closure can reflect the fatigue state of the driver under certain conditions,and the external shape of the human eye conforms to the elliptical feature.Therefore,after positioning to the human eye,this paper selects the elliptical fitting method to extract the extracted eye contour.The information is fitted.The ellipse fitting of the least square method can fit the equation parameters by establishing the elliptic equation,and the ratio of the ellipse length and the short axis can reflect the degree of human eye closure.After determining the state of opening and closing of the human eye,the P80 criterion in the PERCLOS algorithm and the blinking frequency are selected to determine whether the driver is fatigued.Finally,this paper tests the LFW face database image and the captured video image through experiments.The results show that the algorithm has high accuracy and fast detection speed.However,there are still some shortcomings in the face deflection angle and eye shadow,and the next improvement suggestions are put forward.