Research And Application Of Eye Tracking Based On Image Sequence

Eye tracking technology is mainly by the computer and sensor equipment to collect and analyze eye movement characteristics,and then get the corresponding eye movement trajectory.In recent years,with the development of computer technology and the upgrading of electronic hardware equipment,eye tracking technology has also been booming.More and more researchers apply the technology to fatigue driving,advertising,psychology,virtual reality,military and other fields.Pupil center and corner are very important eye movement features in eye tracking technology.How to extract these eye movement features accurately and quickly has very important research value and significance.In this paper,the location of pupil center and eye corner are studied(1)Aiming at the problems of eyelid occlusion,eyelash occlusion and white spot interference caused by corneal reflection in infrared image pupil center location,this paper proposes a pupil center location scheme based on Star ray model.Firstly,the infrared image of human eye is preprocessed by noise reduction.Secondly,the pupil region is detected by yolov3 deep learning model to realize the rough location of pupil center.The star ray model is used to extract the pupil edge points,and the error of the pupil edge points is analyzed to eliminate the false pupil edge points.The remaining edge points are fitted by the least square method,and the ellipse is modified to locate the pupil center accurately.The experimental results show that this scheme improves the accuracy of pupil center location and has good real-time performance.(2)Aiming at the problem of corner detection in infrared image,such as illumination and eyelash occlusion,this paper proposes a corner detection scheme based on random forest.Firstly,the eye region is detected by yolov3 deep learning model.Secondly,the eyes are binarized by Otsu.On this basis,the upper eyelid points are extracted by scanning line algorithm;In this paper,the machine learning algorithm based on random forest is used to extract the lower eyelid points.The upper and lower eyelid points are fitted into a curve by polynomial,and the intersection of the curve is the inner and outer corner points.Experimental results show that the scheme has good real-time performance and robustness.(3)Based on the extracted pupil center and corner points,an eye movement analysis system is built.The system first extracts the center point of the eye pupil and the inner and outer corner points of the eye.By analyzing the position relationship between the two points and the position change of the pupil,it can judge the action of the eye,including moving right,moving left,moving up,moving down and staring,and drive the 3D model to do the corresponding action.