From physics-based representation to functional modeling of highly complex objects

Abstract

The physics-based modeling paradigm augments standard geometric representations with the principles of physical dynamics. It yields powerful models that unify the representation of object shape (geometry) and motion (dynamics) within a single computational framework. Thus, physics-based object representation transforms abstract geometry into real world, object-oriented Geometry++ with potentially enormous benefits for computer vision. In this paper, I will first review some of the physics-based models for vision that we have developed in recent years, including deformable models, physics-based recursive estimators, and dynamic splines, plus some applications to medical image analysis and CAGD. I will then preview a promising future direction for the physics-based modeling approach — the functional simulation of complex, living things and the use of sophisticated models of animals as virtual robots for the synthesis of active vision systems.