Hierarchical decentralized fusion from correlated sensor measurements

In this paper we consider hierarchical decentralized fusion of possibly correlated noisy measurements of a random variable. Our goal is to obtain initial estimates in a decentralized fusion (based on disjoint groupings of the measurements) so that, when these estimates are fused, they give a good overall estimate. In general, this final estimate is worse than the one based on all measurements, this decentralized structure, however, has other advantages that can potentially outweigh this compromise in performance. Since most works on multisensor data fusion assume that noise among different sensors is uncorrelated (i.e.,the noise covariance matrix has a block diagonal structure) which is not always a valid assumption, our approach in this paper allows us to analyze the degradation in performance incurred when we erroneously assume uncorrelated sensor measurements. With the help of sensitivity analysis, upper bounds on this degradation are derived in terms of the off-block diagonal part of the noise covariance matrix that is not taken into account.