Evidential Networks for Decision Support in Surveillance Systems

This chapter focuses on valuation algebras in the context of the theory of evidence, due to the expressive power of belief functions that can represent both classical probability functions and possibility/necessity functions. It introduces the concept of valuation algebra for knowledge representation and reasoning under uncertainty. The chapter describes the algorithms for local computation in a valuation algebra: the fusion algorithm, the binary join tree (BJT), and inward propagation. It presents the basic tools of the theory of evidence as a valuation algebra: the belief functions, the combination and marginalization operations, and the approaches to decision making. The chapter illustrates the theoretical concepts by two examples: decision support systems for target identification and threat assessment. A BJT can be seen as a data structure that allows the intermediate results of the combination process to be saved and the marginals to be computed efficiently.