A Spatio-Temporal Context Tracking Method Blended with LBP Texture Feature

Abstract

A spatio-temporal context (STC) tracking method blended with LBP texture feature is proposed with regard to the problem that it's hard to track a target effectively by a conventional STC tracking method where target background mutation, shading or shape change takes place during tracking. First, the similarity in the texture histogram of the target area between the first image frame and each image frame behind it is calculated to solve the problem of central position shift as a result of the background mutation of the tracked target. The similarity in the texture histogram of the target area between two adjacent frames is then calculated to solve the problem of shading that happens to the tracked target. Finally, a judgment is made as to whether or not the STC update coefficient is changed and the central position coordinates of the tracked target offset, based on the relation between the worked-out similarities in texture histogram and the threshold. Shape change does not alter local texture information about the tracked target and the STC tracking method blended with LBP textural feature has thus good robustness to shape change. Experimental results show that the method in this paper sees a 73.06% increase in average rate of successful tracking and an 89.94-pixel decrease in average error in target's central position, compared with the original STC tracking method. The method in this paper allows more stable tracking of a target where target background mutation, shading or shape change takes place.