Why Multi-Source, Multi-Target Data Fusion is Tricky
Abstract
The purpose of this paper is to demonstrate that, when the number of targets is not known beforehand, Bayesian optimal filtering approaches to multisensor-multitarget data fusion problems encounter unexpected conceptual and practical difficulties. The reason is that single-target Bayesian filtering cannot be naively generalized to multitarget situations and that, consequently, serious pitfalls await those who simply "declare victory." In particular, we show that the classical Bayesian techniques for optimally determining parameters of interest--e.g., the maximum a posteriori (MAP) and expected a posterior (EAP) estimators--cannot even be defined in multitarget situations. We describe our solution to this problem, "finite-set statistics" (FISST), as well as "joint multitarget probabilities (JMP)," a renaming of a special case of FISST. We show how FISST leads to provably Bayes-optimal multisensor-multitarget data fusion algorithms. We discuss the optimality and convergence properties of two different Bayesian data fusion algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 07, 1999
- Accession Number
- ADA392200
Entities
People
- Ronald P. Mahler
Organizations
- Lockheed Martin