Applications of hybrid system identification in computer vision

In this expository paper we illustrate the use of filtering and identification-theoretic techniques in a number of problems in computer vision. We first demonstrate how linear system identification techniques combined with distances for linear systems can be used for modeling, synthesis, classification and recognition of dynamic textures and human gaits. We then show how hybrid system identification techniques can be used for segmentation of dynamic textures. We also highlight some open problems in system identification that are motivated by extensions of the research described in this paper.