NONLINEAR FILTERING - MONTE-CARLO PARTICLE RESOLUTION

The optimal filtering problem concerns the estimation of a process from its noisy partial observation. With the notable exception of the linear-gaussian situation, general optimal filters have no finitely recursive solution. The particle resolution procedure introduced in this paper offers an alternative to local or a priori approximation. II is a global measure-approximation made on the probability space. Assuming some natural regularity and detectability conditions, the particle filter converges to the optimal filter uniformly with respect to time. It has been applied to several technological problems, with unstable systems ([2] to [5]).