Comparison between gaze and moving objects in videos for smooth pursuit eye movement evaluation

Abstract: When viewing moving objects in videos the movement of the eyes is called smooth pursuit. For evaluating the relationship of eye tracking data to the moving objects, the objects in the videos need to be detected and tracked. In the first part of this thesis, a method for detecting and tracking of moving objects in videos is developed. The method mainly consists of a modified version of the Gaussian mixture model, The Tracking feature point method, a modified version of the Mean shift algorithm, Matlabs function bwlabel and a set of new developed methods. The performance of the method is highest when the background is static and the objects differ in colour from the background. The false detection rate increases, when the video environment becomes more dynamic and complex. In the second part of this thesis the distance between the point of gaze and the moving objects centre point is calculated. The eyes may not always follow the centre position of an object, but rather some other part of the object. Therefore, the method gives more satisfactory result when the objects are small.