Nonadditive probability, finite-set statistics, and information fusion

Information fusion is the name given to military expert-systems problems. In this paper we summarize recent work proposing a fully probabilistic theoretical unification for much of information fusion based on the theory of random sets. Our approach unifies detection, localization, classification, and prior knowledge with respect to these. It also unifies precise data together with imprecise data and propositional or vague/fuzzy evidence, as well as certain associated methodologies (e.g., fuzzy logic, rules). Underlying our approach is the discovery that classical single-sensor, single-target point-variate statistics can be directly generalized to a multisensor, multitarget statistics of finite-set variates. We describe “finite-set statistics” and its application to multisensor estimation using diverse data forms. We also point out relationships with current theoretical statistics