Generalized chernoff fusion approximation for practical distributed data fusion

This paper advances research in practical distributed data fusion with an emphasis on the generalized fusion of probability density functions in the presence of unknown correlations. Specifically, the proposed algorithm addresses fusion of any finite number of probability density functions in a distributed tracking environment where “rumor propagation” and statistical correlations may be present. This “rumor propagation” arises in real-world tactical military applications where distributed fusion nodes have dynamic and multi-cyclic data flows. In addition, interoperability requirements with legacy systems preclude control over pre-processing of data fusion inputs to ensure statistical independence or modify legacy systems with pedigree tagging techniques. Leveraging the well-known Covariance Intersection algorithm, its generalization, and previously developed approximations to Covariance Intersection, a computationally simple approximation for the generalized fusion of any number of probability density functions is presented as the novel result of this paper. The derivation of this algorithm and numerical examples illustrate that the proposed approach enables practical fusion of generalized (non-Gaussian) observations in an ad-hoc distributed fusion network without the need for pedigree tagging.