Study On Gaze Tracking And 3D Point-of-regard Measurement Method Based On Spatial Pupillary Coordinate

Free-space point-of-regard estimation is an advanced application of the gaze tracking method,and one important research aera is to apply it to the measurement of the size of the object being watched.According to the application of point-of-regard estimation in free space based on gaze tracking technology,and considering the current research status of the gaze tracking method,the key theories and research topics of this study are,1.Improving the accuracy of binocular line-of-sight estimation.2.Achieving 3D point-of-regard estimation in free space without scene information.The study of this paper mainly focuses on the theory and technology of gaze estimation and 3D point-of-regard estimation.Therefore,a method of simplified 3D eye model for line-of-sight estimation and a free-space point-of-regard estimation method based on the prediction of gaze depth information are proposed.Two common 3D geometric model-based line-of-sight estimation methods,3D eyeball-based method and iris/pupil contour-based method,are chosen to analyze.The geometric models and the algorithms for calculating line-of-gaze/sight are all presented,and the calculation errors are also discussed in detail.Based on the analysis of the existing methods,the paper studies the theoretical models,the calculation of the visual axes and the estimation methods of 3D point-of-regard in free space,and proposes two theoretical methods for estimating 3D point-of-regard in free space and a system calibration method for the proposed stereo vision-based gaze tracking system.The main research works are as follows:First,the paper proposes a simplified 3D eyeball model-based line-of-sight estimation method.This method retains the pupil center,eyeball center,inner eye corner and the angle between the optical axis and the visual axis as four kinds of eye features for calculating the parameters of the visual axis.The advantage of this method is that reducing the personal calibration process while keeping enough accuracy of the calculated visual axis.The method outputs the parameters of binocular spatial line-of-sight with high accuracy.The mean error and the root mean square error of line-of-sight estimation are 0.98 ± 0.45 degree,when watching a calibration board with a distance of 2 meters.Second,the paper proposes a 3D point-of-regard estimation method which combines the vector from pupils' midpoint to real point-of-regard and Gaussian process regression algorithm.Specifically,Gaussian process regression is used to build the mapping between the depth information of point-of-regard and the eyes' features parameters which including binocular pupil centers and the line-of-sight.This method reduces the estimation error of the depth information of point-of-regard using the data extracting from personal calibration process.The results of the experiments of watching a calibration board in different distances show that the estimated point-of-regard accuracy by using the proposed method is better than the result obtained by the previous 3D point-of-regard estimation model,and the mean error and the root mean square error of the estimated point-of-regard are 15.41 ± 11.78 centimeters when watching objects within 2 meters.The paper also tests the proposed method's performance when used in real scenes by using cubic gauge blocks.Third,a stereo vision-based gaze tracking system is proposed for achieving the above two main methods.And the paper studies the system calibration method and the pupil contour extraction algorithm for stereo cameras.