Methodes de filtrage pour du suivi dans des sequences d'images - Application au suivi de points caracteristiques
Élise Arnaud · HAL (Le Centre pour la Communication Scientifique Directe) · 2004
This thesis is concerned with the use of filtering methods for tracking in image sequences. For the algorithms introduced here, the system is represented by a Hidden Markov Chain, described by a dynamic law and a likelihood. In order to construct a general method, the dynamic law is estimated from the images. This choice underlines some limitations of the simple model of Hidden Markov Chains. Indeed, such a modelisation does not describe the dependance of the system's components to the image sequence. We first propose an original modelisation of the problem where the image data are explicitely taken into account. Such a model allows us to consider algorithms that do not rely on a priori information. Different kinds of filters associated with this new modelisation are derived. Then, a validation of this modelisation is presented. Three feature point trakers are proposed on this basis. They combine a dynamical law relying on an estimated motion, and measurements provided by a matching technique. Finally, this approach is extended to planar objet tracking.