Research On Head-Mounted Gaze Tracking Algorithm

The human eye is the most important way to obtain information in daily life,and gaze tracking is an application technology that uses human eye movement information to obtain gaze direction or attention,and has a wide range of applications in safe driving,human-computer interaction and other fields.Sight tracking system according to its application form can be divided into desktop and head-mounted two,compared with desktop system,head-mounted system system is theoretically not limited by a specific range,has a broader application prospects,is the current line of sight tracking research hotspot.This article is divided into two parts to investigate the head-mounted gaze tracking algorithm.The pupil center positioning section addresses the poor robustness of pupil positioning under non-ideal conditions,an improved OTSU threshold segmentation and RANSAC ellipse-fitting pupil localization algorithm is proposed,and the positioning idea from coarse to fine is used to locate the pupil of the human eye.Firstly,the acquired images are preprocessed,including RGB to grayscale map,image filtering,and morphological processing,which provide good conditions for the subsequent steps.Then,according to the grayscale characteristics of the human eye image,a modified OTSU threshold segmentation algorithm with optimized grayscale search interval and weighted threshold is used to perform threshold adaptive segmentation of pupil pictures under non-ideal conditions,and finally the edge points are detected and the pupil ellipse is fitted using the highly robust RANSAC algorithm.In addition,for the occlusion problem,whether it is occlusion is judged according to the Papanict distance,and the previous frame parameter is used to compensate for the occlusion image.The line-of-sight estimation part adopts the principle of constant intersection ratio to calculate the line-of-sight landing point,which aims at the systematic error of the algorithm model and some problems of the conventional error compensation method,a line-of-sight calculation method based on fixation trajectory compensation is proposed.Firstly,the calculated fixation point is corrected once according to the shape information of the trajectory,and then the method of compensation table is established,the correlation coefficient of the compensation table is solved by the least squares method,and the fixation point is compensated again by using the solved compensation table.Under the research conditions of this paper,the proposed method and the nine-point method are experimented with different calibration screens at the same time,and the error of the fixation trajectory method is significantly reduced compared with the nine-point method and the error distance can converge within15 pixel,which verifies the effectiveness of the proposed algorithm.