Evidence Combination in an Environment With Heterogeneous Sources

Abstract

A framework for the combination of evidence in an environment where data are generated from heterogeneous sources possessing partial or incomplete knowledge about the global network scenario is presented. The approach taken is based on the conditional belief and plausibility notions in Dempster-Shafer evidence theory that allow one to condition these partial knowledge bases so that only that portion of the incoming evidence that is relevant is utilized for updating an existing knowledge base. The strategy proposed enables one to accommodate some of the most challenging, yet essential, features that are encountered when evidence is generated from possibly a large numbers of sources. These include heterogeneity and reliability of incoming evidence, inertia and integrity of evidence already gathered, and potentially limited resources at the nodes where evidence updating is carried out. The proposed framework is applied in robot map discovery using ultrasonic sensors and a real-world scenario where sensor data generated by heterogeneous sensors are used for potential threat carrier-type detection