Non-contact Eye Gaze Tracking Technology Based On Cornea Reflection

The eye gaze tracking technology has great effect in studying people’sintention, needs, cognition process etc. and can be used in many aspects, such asHCI (assistive devices for disabled), psychology (cognition study), industrial andcommercial (ads analytics), etc. According to whether technologies haveinterference to users, eye gaze tracking technologies are divided into two categories:contact and non-contact. Non-contact type is more popular since it is moreconvenient. Most of non-contact technologies utilize infrared illuminators sincecorneal reflection is very helpful for calculation for point of gaze. There exist manywell research results in eye gaze tracking technologies based on corneal reflection.But for some key issues such as pupil extraction, spots extraction, points of gazeestimation, etc. there is still no accurate enough method with well instantaneity. Anovel non-contact eye gaze tracking method is proposed using cornea reflection.5infrared illuminators and1camera are used in the study. The main contribution ofthis study is as follow:The gray scale distribution feature is utilized to fast location of eye area. Sincethe current methods are lack of accuracy and instantaneity in the eye location stage,the character that the pupil area is the darkest part of the eye image is utilized tolocate the eye area roughly.A new feature spots search method is proposed using BFS algorithm, geometryfeature of the eye image and the gray scale. The new method can overtake thedrawbacks of traditional feature spots search methods using dilation and erosion byextracting the correct spots and match the spots with the illuminators accurately.A cycle Least Square pupil fitting method is suggested with one-dimensionaledge detection. The traditional methods using Sobel and Canny algorithms may losesome actual contour points. The proposed method is run circularly until a stablepupil center is produced. The suggested method can yield more robust results thantraditional methods using Sobel or Canny algorithm.A dynamic calibration method is proposed to overtake the problem that currentspace mapping methods may have overall bias and loss of accuracy. The screen areais divided into small regions, and parameter is extracted for each sub-region. A parameter matrix is extracted for each user. The best parameter is selecteddynamically from a matrix in gazing procedure.Tow group of experiments are done. The first group is accurate test experimentthat the subjects are required to gaze4*4calibration points in the screen. Under theexperiment environment the errors are less than30pixel radius. The second group isapplication experiment, attention analysis and gaze manipulation experiments aredone to verify the usage potential of the proposed method.