Research On Auto-calibration Gaze Estimation Method

At present,eye tracking technology has become a hot topic in the field of visual interaction.It is a technology to obtain the subject’s current three-dimensional line of sight direction or line of sight location in real time,objectively and accurately by using mechanical,electronic,optical and other detection means.Aiming at the related difficulties of eye tracking technology,this paper aims to study two aspects: eye movement feature extraction algorithm under near-infrared illumination and line of sight estimation model based on this feature.1)The existing pupil detection method is improved,and the performance of the existing pupil detection algorithm is improved by using the cascade Haar feature and Otsu dynamic threshold segmentation.The algorithm is simple to implement and has high accuracy.The algorithm is compared with the classical Starburst,Swirski,Ex Cu Se and Else algorithms in the same data set.The results show that the accuracy of the improved algorithm is 86.33% and 89.67% respectively within the error range of 5 pixels and 10 pixels.Compared with Starburst,Swirski,Ex Cu Se and Else algorithms,the detection rate of the algorithm within the error range of 5 pixels is increased by 63.53%,5.63%,6.13% and 10.63% respectively,indicating that the algorithm has better accuracy,detection rate and robustness.2)Based on the analysis of the existing two-dimensional and three-dimensional line of sight estimation models,a self calibration line of sight estimation method is proposed.In this method,firstly,the optical axis direction of both eyes is obtained by using the contour based method,and then the estimated visual axis direction of both eyes is calculated by using the horizontal and vertical angles between the optical axis and the visual axis in the average value table of human eye parameters.The estimated landing point of the visual axis of both eyes on the measured plane is obtained,and the midpoint of the visual axis of the two eyes on the landing point of the measured plane is taken as the final line of sight estimation point of the human eye on the measured plane.The minimum number of cameras and light sources can be a single camera and a single light source.The system complexity is low.The user can estimate the position of the fixation point without complicated calibration procedures under free head movement,meeting the user’s requirements for comfort and complexity.The average angle gaze error of this method is 1.556°.