Visual tracking of a stochastically mobile object

In this paper, a model-based image recognition method is studied, which can track an object with an unpredictable motion and changeable image shapes by using genetic algorithm searching techniques on the unprocessed raw images. We take the advantages of genetic algorithm´s global and parallel searching ability by coding the target´s image information into genes of genetic algorithm´s chromosomes. Some context information of the target, such as neighborhood principle between consecutive image frames, is also used to enhance the matching algorithm´s robustness and reduce its computation time. The feather tracking practices show that this method works