Gaze Estimation System Based Oil Image Texture And Support Vector Regression

The eyes are the main channels for external information into the brain, the study ofwhich helps to understand the process of information processing. With the rapiddevelopment of computer science and biological medicine, in the absence ofinterference environment, eye tracking technology which uses the computer to theperson’s visual process has become one of the highlights in agro-scientific research inthe intelligent man-machine interactive way.This thesis first expounds the background of the research, introduces thedevelopment status of eye tracking technology at home and abroad and the requirementof eye tracking system in the future, then discusses the eye tracking technology tosolve the problemv in accordance with the present situation of the application of eyetracking technology in many fields.Next, from the physiological structure of the eyeball, this thesis introduces theexisting fixation point estimation method of eye tracking systems and eye calibrationbased on3d model and calibration based on two-dimensional regression. Contraposingeye tracking calibration function modeling, different equations are used to studyregression equations based on pupil-corneal vector to fit points of fixation. Becausethe solution of the calibration function is actually a small sample and nonlinearregression problems, this thesis puts forward the SVR algorithm based pupil-cornealvector to estimate the user’s gaze direction.Eye images are performde histogram statistical wirh different kinds of LBP basedon image texture feature theory, the simulation result shows that LBP histogramstatistics can characterized the differences of eye images to express the informationabout the direction of gaze. On this basis, to reduce the dimension of the texture featurevector and improve the ability to extract more texture feature, histogram feature vectorswhich is extracted with feature extraction algorithm based on multistage CS-LBP, isused as the input into the SVR to estimate the user’s gaze direction.The thesis finally proposes the eye tracking fixation point estimation algorithmbased on feature fusion according to the information fusion theory. After discussingmodel feature and texture feature selection of eye images, the pupil-corneal vector andmultistage CS-LBP histogram feature vector are fused as the input feature vector intothe SVR to estimate the user’s gaze direction. Finally, the fusion feature is feasibile togaze point estimation which is verified by experiment. And compared with the SVR gaze point estimation based on single feature, the experiment results show that eyetracking fixation point estimation algorithm based on feature fusion has obviousadvantages over any other individual characteristics as the input into the SVR in themean square error, square correlation coefficient, and the precision.