Data fusion algorithms for distributed nonlinear estimation and tracking

Data fusion algorithms are presented for several nonlinear estimators, used in a distributed sensor network. The data fusion algorithms combine estimates from local nonlinear estimators in order to obtain an improved global estimate. A new algorithm for tracking maneuvering targets, called the Probabilistic Maneuver Association Filter (PMAF), is introduced and its results are compared to those of a previous filter designed for tracking maneuvering targets. A data fusion algorithm for the PMAF is presented, for both communication at every sample interval, and for infrequent communication.; The previously proposed Gaussian sum filter is applied to the bearings-only tracking problem. The Gaussian sum filter recursively updates the a posteriori density function by approximating the density function by a convex combination of Gaussian density functions. A data fusion algorithm is presented for the distributed Gaussian sum filter. The Cramer-Rao lower bound, which is a lower bound on the error of an unbiased estimator, is given for the distributed bearings-only tracking problem. The results of the data fusion algorithm are compared to the Cramer-Rao lower bound in order to measure the performance of the algorithm. The Gaussian sum filter is compared to an extended Kalman filter at both the local processors and the data fusion center.; Factorized forms of the PMAF and the Probabilistic Data Association Filter (PDAF) are presented. These formulations of the PMAF and PDAF are less sensitive to the effects of numerical roundoff error, which result from using finite precision arithmetic. The number of floating-point operations for both the standard and the factorized forms of these filters are given so that one can evaluate the computational requirements of both forms.