Research On Pupil Localization Algorithm Based On Video Images

With the development of technologies in the field of human-computer interaction,the researchers have paid attention to the eye tracking technology.Vision is the major channel to environmental information,perception and understanding of the objective world for people.Human visual attention position in the three-dimensional space is called the gaze of point.Its changes reflect the changes of people's attention,interest and emotion.Pupil localization is an important part of the gaze research.The issues such as the pupil distortion and many objective disturbances restrict the localization accuracy,thus further affect the system accuracy and stability.To settle the problem of how to quickly and effectively eliminate the interference noise and improve the accuracy of the pupil localization,we propose an algorithm of pupil localization combining area voting and model constraint with the information of the feature point location and gradient.The main contents list as follows.Section one,the detection and localization of the eye is the preliminary step of the pupil localization.We use the Adaboost algorithm to detect the face in the video image.Then the eye areas are located with the eye position in the face.This step is the foundation of the pupil localization.Section two,the algorithm for pupil localization combining area voting and model constraint is elaborated.Firstly,it extracts the edge points with the modified Starburst algorithm,and votes in the area based on the gradient feature.Then it finds out the centroid of threshold area with high voting values,and calculates the distances between the edge points and the centroid.And it extracts inliers by analyzing the statistical feature of those distances.Secondly,it refines inliers by the model constraint iteratively and finds the pupil location.Section three,the pupil of image on the eye database is located with the algorithm proposed.With the aspects of the detection accuracy and time complexity of the algorithm,we compare this algorithm with three algorithm for pupil localization.Then we analyze the advantages and disadvantages of the algorithms in detail.Section four,it describes the pupil detection platform,which includes the hardware system and software system,especially the open-source and cross-platform image debugger visualizer design for program debugging.Experiment shows that this algorithm not only has strong robustness,but also has a good localization in case of the distortion pupil and occlusion.Apart from that,it is fast for real time application and satisfies the accuracy requirement of gaze system.