Filters for estimating Markov modulated Poisson processes and image-based tracking (Q1360456)

The paper studies problems of estimation of parameters and for various statistics of Markov modulated Poisson processes (MMPP), i.e. for doubly stochastic Poisson processes with intensities depending on the state of a continuous time finite Markov chain. Both innovations and Zakai finite dimensional filters are derived for the estimation of the statistics of the MMPP. These filters are then used in an expectation maximization (EM) algorithm which provides maximum-likelihood estimates of the Poisson intensities and of the transition probabilities of the Markov chain. An application to image-based tracking of maneuvering targets is finally illustrated.