A Framework for Generic State Estimation in Computer Vision Applications

Experimenting and building integrated, operational systems in computational vision poses both theoretical and practical challenges, involving methodologies from control theory, statistics, optimization, computer graphics, and interaction. Consequently, a control and communication structure is needed to model typical computer vision applications and a flexible architecture is necessary to combine the above mentioned methodologies in an effective implementation. In this paper, we propose a three-layer computer vision framework that offers: a) an application model able to cover a large class of vision applications; b) an architecture that maps this model to modular, flexible and extensible components by means of object-oriented and dataflow mechanisms; and c) a concrete software implementation of the above that allows construction of interactive vision applications. We illustrate how a variety of vision techniques and approaches can be modeled by the proposed framework and we present several complex, application oriented, experimental results.