An Algorithm Of Pupil Center Tracking Based On Optical Flow

Some important information of the human brain and even the physical condition can be obtained or revealed through eye movement information. While pupil tracking is an important part of eye tracking, one of the important methods of analyzing brain status and physical behavior. Research on pupil tracking technology has been a hot topic, attracting people’s attention. And it is an important topic in the field of computer vision, involving computer vision, image processing, psychology, physiology and ergonomics and other fields, which is rising with the development of face detection and analysis, detection and analysis of expression. The research achievements in the military, commercial, industrial and other fields have been widely recognized.The commonly used pupil tracking algorithms include nonlinear filtering theory, pattern and template matching method, etc. While the method based on template matching, only after precise location of eye, performs pattern matching according to the geometric features of the pupil, so as to realize the tracking. However, this method is seriously influenced by disturbance environment, it is difficult to guarantee the robustness of the method and accuracy.Firstly, this paper studies tracking algorithms related in detail, summarizes the advantages and disadvantages of each algorithm, and then proposes a pupil center tracking algorithm based on optical flow method. This paper briefly introduces the basic principle related at first, and models the abstraction of pupil plane image. Through the de-noising pretreatment of original image, the pupil image with high quality is obtained. Then Adaboost algorithm is used for training HAAR feature classifier to extract the pupil characteristics of the region of interest. The pupil boundary features in video frame are tracked by frame using pyramid optical flow method, the feature points are calculated by means of the basic formula of spherical triangle and the least squares method, finding out the preliminary correspondence between the new frame feature points and the old. Then perform statistical analysis on these relationships and to determine the final correspondence, calibrate the pupil boundary and correct noise. Finally, the pupil boundary characteristic points are fitted. According to the results of the fitting calculation pupil center, so as to achieve the purpose of pupil center tracking.The pupil center tracking algorithm we proposed in this paper, of which the experimental results compared with the current popular algorithm indicate that the false recognition rate and false rejection rate are reduced, improves the recognition accuracy of the field of the algorithm.