The Research Of Fatigue Driving Detection Based On Image Processing

Traffic accidents occur many times in today’s society.Accidents caused by fatigue driving are not rare.Therefore,the state vigorously advocates the importance of fatigue driving monitoring.Existing fatigue driving detection methods include heart rate monitoring,lane departure monitoring,head position monitoring,and steering wheel monitoring.Due to the driving environment of the driver is very restrictive,this method of contact detection cannot play a good role.This situation is largely unable to achieve the purpose of preventing fatigue driving.In order to achieve the regulatory effect without affecting the driver’s driving.Fatigue driving warning system needs to have real-time and accuracy.Among various methods,the more important one is the monitoring algorithm based on image processing.The thesis is based on a certain image processing technology to study fatigue driving detection.It is mainly detected according to the important fatigue features that the driver will experience during driving.The system comprehensively determines whether the driver is finally in a fatigued driving state based on the PERCLOS value of the driver’s eye area and the human eye curve during the entire driving process.In the detection process of the system,it mainly includes the preprocessing of the picture,the positioning of the face,the positioning of the key points of the human eye,the human eye state,the detection calculation of the PERCLOS value and the drawing of the human eye curve.The main work completed in the topic are:(1)For the fatigue driving warning system,the experiment builds a database of fatigue driving faces that meets the needs of the experiment.Because the existing database with manual annotations does not completely contain facial images that fit the fatigue driving research information.Therefore,the experiment uses a part of the existing facial images in the database with manual annotation.In the experiment,360 students are additionally collected facial data.Each student collects 8 images that meet the experimental requirements,and adds 2880 images for fatigue driving experiments.The database finally adopted by the experiment is highly targeted.(2)Research and implement the image enhancement algorithm.Image enhancement algorithms based on morphology are mainly used in image preprocessing experiments.Another aspect of image preprocessing is the illumination compensation of the image.Illumination compensation and image enhancement can improve the brightness and clarity of the image.After image preprocessing,the subsequent face detection experiments have reduced the false detection rate of faces by 4.3%.(3)Research and implement the face detection algorithm.The face detection method mainly uses the fast recognition feature of the Ada Boost algorithm,and applies the recognized face to the CLM local model.The running speed of the final single image of the system is 46 ms / frame,and the detection time of the image area is reduced by 15%.(4)Research and implement human eye positioning algorithm and human eye tracking.Human eye positioning adopts gray integral projection combined with CLM key point positioning method.According to the horizontal projection and vertical projection of the open and closed state of the human eye,the approximate eyebrow area is determined,and then the CLM is used for secondary positioning.To a certain extent,the calculation complexity is reduced.The final human eye positioning accuracy rate reached 97.8%,the single image running speed was 39 ms / frame,and the overall detection speed of the system was increased by 12%.(5)Research and implement the fatigue judgment algorithm.The human eye curve change method is combined with the PERCLOS judgment criterion.The final fatigue driving detection process can achieve corresponding reliability and accuracy.For 56 specimens with glasses and 73 specimens without glasses,the accuracy rates of fatigue state determination were 92.86% and 94.52%.