Driver’s Eye Movement Detection Based On Apparent Features

With the continuous increase of traffic accidents these years,the development of vehicle-mounted equipment based on the driving status has gradually become a research hotspot in the field of automotive active safety.Due to the strong correlation between the eye state and the driving state,related research based on the eye movement state has attracted widespread attention in aspects like driving fatigue and HMI.Given the technical advantages of machine vision,the detection of the driver’s eye movement based on visual images has become the main technological approach for researching related vehicle-mounted equipment.However,since visual images are susceptible to environmental factors such as light changing,especially for driving conditions,some problems like the low effective resolution of the driver’s eye images and changes in lighting effect need to be resolved in conjunction with actual engineering applications.Accordingly,the paper aims to improve the accuracy of eye movement detection in vehicle application,and puts forward an efficient detection method on eye movement.The main work contents are as belows:(1)Eye detection based on Adaboost algorithm.In view of the problem of high rate of undetected error and misrecognition,this paper uses the Adaboost algorithm to detect the eyes based on the ROI area of the face,which reduces the detection range to improve detection efficiency and reduce the risk of misidentification.(2)Detection of eye’s apparent features.Given the problem of low quality of images due to the small proportion of eyes in the face,this paper aims to enhance the image quality of the eye by superpixel reconstruction technology based on CNN,and then detect the apparent features like pupil and eyelid and tracking them.(3)Construction and verification of eye movement detection model.Based on the research objective,a driver eye movement state(PERCLOS,gaze direction)detection model is proposed combined with the detection results of apparent features,and verified and analyzed by reference data of eye tracker.The detection method on driver’s eye movement state is studied in depth in response to the problems of it,the main breakthroughs are: SRCNN is used to solve the problem of insufficient pixels of eye images;An efficient eye movement detection model is proposed to estimate the gaze direction according to the actual requirements of vehicle applications.