ABSTRACT
A decentralized data fusion (DDF) system is a general, flexible system for fusing information in a robust, distributed, and scalable manner. A DDF network is composed of a set of connected processing centers or nodes. Each node attempts to estimate the underlying state of a system of interest. Nodes can use information that comes from two sources: local or remote. This chapter presents the main principles, challenges, and solution strategies associated with non-Gaussian exact DDF and "rumor robust" conservative DDF. Two measures of information content in the global network are used to assess the consistency of the exact DDF solution: the association probability metric (APM) and the cumulative entropy metric (CEM). Random finite set (RFS) techniques have generated considerable interest for a wide range of multi-sensor fusion applications, especially those involving multitarget tracking. Probabilistic graphical models (PGMs) encode high-dimensional joint probability distributions as products of conditionally independent "factors" that are functions of subsets of random variables.
