Visual tracking of targets in unrestricted environments

Tracking is a basic process for security and surveillance problems that need to follow an object of interest during a certain period of time. Visual tracking algorithms based on pattern matching techniques constitute the basis of many, and the most used, current tracking systems (Brown, 1992). However, the main drawback of template matching consists in its lack of automatic adaptation to any environmental condition. Usually, the object of interest or the environment change their visual aspect and the template should, in a robust way, adapt to them. Therefore, a proper updating technique of the template is a crucial matter of all these systems. This paper proposes a new approach to the updating problem in order to achieve a better tracking performance and robustness. This is carried out by using an internal representation technique that makes use of second order isomorphisms (Shepard and Chipman, 1970). This allows to establish a representation space where an object of interest can be more easily distinguished from the representations of the objects of the context. The most important improvements of this approach are its parameter-free working, therefore no parameters have to be set manually in order to tune the process, and a better performance in robustness compared with other methods. Besides, objects to be tracked can be rigid or deformable, the system is adapted automatically to any situation