Condensation-Based Contour Tracking with Sobolev Smoothness Priors

This paper proposes a combination of contour deformation modelling in Sobolev spaces and the Condensation filter to track an object over a sequence of images. As Sobolev spaces are smoothness spaces this allows to control the smoothness of the contour deformation extending previous wavelet representations. We also introduce a probabilistic model for the wavelet deformation of the contour that induces a prior distribution for contour deformation. The deformation model is used to generate an stochastic dynamic model for contour evolution in time. Computational results are presented that show applications of this formulation.