The role of dynamics in computer vision and image processing

Dynamic vision and imaging systems can substantially improve our quality of life. However, key issues that must be addressed in order to realize this potential are their fragility when used in unstructured environments and their need to process vast amounts of data in real time. The realization that most actionable information embedded in imaging data can be compactly encapsulated in dynamic models of relatively low order provides a powerful insight to address these issues. As we show in this paper, system theoretic tools allows to recast a wide range of dynamic vision problems into a computationally tractable optimization form. These ideas are illustrated with several applications including tracking, segmentation, and texture analysis/synthesis.