Research On Gaze Tracking Algorithm Based On Binocular Stereo Vision Technology

Gaze tracking is a new human-computer interaction technology which has great development potential,it is more natural,more directly and more human compared with the traditional ways,and has a wide range of applications in game entertainment,helping the old and disabled,smart home,advertising analysis and other fields.The current gaze tracking system is most of user's head movement restrictions,need to use infrared light and other auxiliary equipments to ensure tracking accuracy which is complex and costly.In this thesis,a gaze tracking technology that can overcome the head movement restriction in natural light is studied,and a gaze tracking algorithm based on binocular stereoscopic vision is proposed to track the gaze real-timely and accurately which calculates the coordinate of the fixiation combined eye movement characteristics extracted in natural light with the head posture correction.The main research contents of the thesis are as follows:(1)Eye movement features extraction in natural light.Firstly detect and track the face from video image,use a method combined with distribution of facial features and human eye's gray features which is first rough and then accurate to split the eye area.An iris center localization algorithm based on circular operator and gray projection is proposed for the low contrast and nonuniformity of the human eye region image in natural light.Then,the Harris corner detection algorighm is used to extract the inner corner of the eye after the pixel preselection,finally real-time and accurate positioning of the eye movement feature points is realized.(2)Head posture recognition.A head posture model is constructed with the inner corners of left and right eyes and the midpoint of the left and right mouth corner,and a head posture calculation method based on the 3D informations of feature points is proposed.Firstly,edge detection and Harris corner detection algorithm are used to locate left and right mouth corners,and then according to the principle of stereo vision and binocular camera calibration parameters to calculate the three-dimensional coordinates of the feature points,finally,calculate the up and down pitch angle and left and right deflection angle of the head in accordance with the geometric relationship of the three feature points in three-dimensional space,and the head posture is acquired.(3)Fixation point location.Firstly,the gaze point mapping model is established based on the relationship between the eye movement vector and the position of the fixation point,the relevant parameters of the model are obtained by calibration,and then the fixation position is rough calculated according to the detected eye movement vector.Considering the influence of head movement on the fixation point,a method of twodimensional mapping model combined with head movement compensation is proposed,the fixation position is corrected according to the pitch angle and the deflection angle of the head.This thesis verifies the proposed gaze tracking algorithm on VS2010 platform,the results show that the scheme allows the head to move within a certain range while ensuring the positioning accuracy of the fixation point.Subsequently the research group applied the proposed algorithm scheme on the embedded platform,it proves that the proposed algorithm scheme can meet the requirements of tracking system in terms of accuracy,real-time and feasibility.