Intégration de relations spatiales floues dans un filtre particulaire pour le suivi d’objets

In this paper, we propose a novel method to introduce structural spatial information in particle filters. This information, expressed as spatial relations such as orientation and distance, is modeled in a fuzzy set framework, and is introduced in the dynamics in order to model the potential changes from one instant to the next one. The fuzzy modeling provides flexibility both in the semantics of the relations and in the transitions from one relation to another one. This allows taking into account situations where a tracked object changes its direction in a quite abrupt way. Tests on synthetic data show that our approach efficiently tracks objects with complex dynamics, and outperforms several classical filtering techniques while using only a small number of particles.