Template apdating based on context

Abstract

Visual tracking based on pattern matching is a very used computer vision technique in a wide range of applications [3]. Updating the template of reference is a crucial aspect for a correct working of this kind of algorithms. This work proposes a new approach to the updating problem in order to achieve a better performance and robustness of tracking. The proposed technique has been compared experimentally with other existing approaches with excellent results. The most important improvements of this approach is its parameter-free working, therefore no parameters have to be set up manually in order to tune the process. Besides , objects to be tracked can be rigid or de-formable, the system is adapted automatic and robustly to any situation. An added memory of patterns allows to improve the robustness. 1. Introduction Visual tracking based on pattern matching is a very used computer vision technique in a wide range of applications [3]. Its working is simple, a template of reference is searched in the current image. Due to that the view of the tracked object can modify its aspect through the time, an update of the template is necessary in order to keep the object. Two problems can arise because of the in-sucient or excessive frequency of the number of updates. In the rst of them, the visual aspect of the object of interest can become too dierent from the pattern and, in this way, the searching algorithm can nd other part of the searching window more similar to the current pattern. This produces a jump in the object of interest. The other problem is due to applying too and unnecessary updates to the template of reference. The digital nature of images and patterns can cause drifting due to an accumu-lative sub-pixel error in every update. Sometimes , the random movement of the object can counteract the eect of the drift, but generally the drift can be signicant and end up losing the object. Every update causes a potential drift. This work proposes a new template updating approach within the framework of representation spaces based on second order iso-morphisms. Among its advantages are a parameter free working, no parameters have to be set up manually prior working, and a better performance than the traditional updating methods.