This project aims at developing robust tracking algorithms for dealing with complex motion and shape dynamics. Current methods achieve good tracking performance in the presence of non-cluttered background and smooth motion and deformation regimes but they typically loose track if one of these conditions fails. This project studies the use of multiple deformable models with switching/mixing algorithms as a way to enlarge the tracking capabilities. This raises several interesting questions: how to combine multiple models or switch between them? Can tracking be improved if the parameters lie on manifold contained in Rn? What methods can be used to estimate the manifold and the parameter trajectory? Other topics that will be addressed in this project are multi-model learning and robust trajectory estimation methods to reduce the influence of outliers. The proposed algorithms will be tested in selected applications.