Research Of Sight Tracking Algorithm Based On Pupil-corneal Reflex

As the human eyes have a natural, direct and bilateral access to get informatin, thesight tracking technology based on capturing and processing the pictures of eyes has abroad application prospects and a great business value on Human-computer interaction,medicine, psychology, and military etc, therefore, it has become a hot topic in the fieldof computer vision, arousing more and more researchers to join in it.This paper has studied some correlation algorithms on gaze tracking based on thetheory of the pupil and corneal reflection, the main contents are described as follows:①The eyes detection and position algorithms: it has realized the work ofdetecting eyes from the vedio pictures to avoid the blind searches in the processing ofeyes fixation point’s position, it can be divided into three parts: Human face dection andposition, eyes position, as well as eyes tracking. In the process of human face dectionand position, this paper has adopted the Adaboost based face detecting algorithms aftercomparing a veriety of face detecting algorithms. In the process of eyes positioning, thispaper has proposed an algorithm with combining three kinds of features of eyes: Firstly,it uses a priori knowledge of the human face to positioning eyes region roughly.Secondly, it isolates the eyebrows according the eyes’ gray distribution.finally, it canpositioning the eyes region precisely by extracting the inside corner of each eye. Inaddition, In the process of eyes tracking, this paper has studied an eye trackingalgorithm by tracking the inside corners of eyes with a predicting algorithm based onthe Kalman filter, it can track the eyes when human’s head moving, and when it workswell, it also can raise the speed of the eye detection and position algorithm bydecreasing the size of ORI images.②The eyes fixation point’s detection and position: it has realized the work ofgetting the pupil-facula offset from the eyes subimage and mapping it to the real sceneto extract the position of eye fixation point, it also can be divided into three parts: pupilcenter positioning, facula center positioning and calibration process. In the process ofpupil center location, the paper has introduced an algorithm scheme including imagesegmentation, pupil rough location, pupil edge extraction and edge fitting: Firstly, thispaper has adopted an improved OTSU algorithm to complete the image segmentation.Secondly, it has adopted the centroid algorithm to calculate the pupil center roughly.Thirdly, it has calculated the subgraph’s gradient by a radial scanning with the one-dimensional operator, which can extract the pupil’s edge points accurately. finally,it has chosen the way of edge points’ geometry fitting way to get the position of pupilcenter accurately. In the process of facula center position, this paper has studied a spotcenter position algorithm based on Gauss fitting, which can achieve a high precisionwith an error less than0.1pixels. In the process of calibration, this paper has studied avariety of calibration models, and has chosen the least square method to establish a6parameters mapping function model, which can accurately calculate the eyes fixationpoint in real scenes with the relative offset between the pupil center and the faculaecenter. With the algorithms above, the paper has achieved the purposes of gaze tracking.In this paper, it has simulated the algorithms on Matlab and has tested it in apractical system, and has proved that the algothms have a good Feasibility, accuracy andreal-time performance.