Real-time tracking using A* heuristic search and template updating

Abstract

Many vision problems require fast and accurate tracking of objects in dynamic scenes. In this study, we propose an A* search algorithm through the space of transformations for computing fast target 2D motion. Two features are combined in order to compute efficient motion: (i) Kullback-Leibler measure as heuristic to guide the search process and (ii) incorporation of target dynamics into the search process for computing the most promising search alternatives. The result value of the quality of match computed by the A* search algorithm together with the more common views of the target object are used for verifying template updates. A template will be updated only when the target object has evolved to a transformed shape dissimilar with respect to the actual shape. The study includes experimental evaluations with video streams demonstrating the effectiveness and efficiency for real-time vision based tasks with rigid and deformable objects.