Contour tracking in complex environments using wavelet basis

Abstract

In this paper, a new tracking algorithm for active contours using wavelets is presented. First it is shown how to construct contour descriptions based on a multiscale representation of planar shapes using wavelet basis. Second, in order to model smooth contour transitions between image frames, probabilistic shape priors for modelling contour deformation using wavelets are presented. And finally this new formulation is applied to the problem of tracking a contour in a cluttered environment using stochastic models to predict contour location and appearance in successive image frames. These three components are integrated in the Condensation (Conditional Density Propagation) tracking algorithm which is specially designed to work in cluttered environments. Computational results are given for two real image problems and show that this formulation successfuUy tracks the objects in the image sequences.