Multisensor fusion for target tracking using sequential Monte Carlo methods

In this paper, we consider the problems of centralized and distributed multisensor filtering from a Bayesian perspective. We present sequential Monte Carlo algorithms for obtaining complete posterior distributions from individual sensor measurements and from individual sensor posterior distributions, respectively. In the latter case, the individual posterior distributions are approximated as Gaussian distributions, where the information being communicated by the sensors are the statistics of the distributions. The posterior distributions obtained by a centralized algorithm are computed either by the fusing of the likelihoods or by combining the moments of the individual sensor posterior distributions. The proposed algorithms are applied to two problems of target tracking (a) using bearings only measurements and (b) using multimodal sensor data. For the problems, we provide the root mean square errors, and for problem (a), we compare them with the posterior Cramer-Rao lower bounds.