Source Conflicts in Bayesian Identification

In Bayesian identification an ID source is in conflict with the other ID sources, if both provide substantially different, reliable information on a tracked object. After discussing some general aspects of source conflicts and introducing two established conflict-definition approaches, it is denoted that these approaches each show a counterintuitive effect. By applying a conflict definition from the theory of Bayesian networks, the Coherence Approach and as refinement the Extended Coherence Approach are proposed. In an experimental evaluation, all approaches are compared with each other and with the expert’s intuitive understanding of source conflicts. 1 Bayesian Identification Identification of a tracked object is the characterization by assignment of an identity (ID). This identity describes object features, e.g., allegiance or intent, necessary to know in performing military missions. The standard identities Friend, Assumed Friend, Neutral, Suspect, Hostile, and Unknown are often used in context of military air surveillance and defense. Depending on the application context, different identities might be defined and applied. Bayesian identification facilitates fusion of uncertain information from manifold ID sources, e.g., track behavior, IFF equipment, Electronic Support Measures, and adherence to air traffic regulations. Subsequently, we sketch an exemplary Bayesian identification process. More details can be found in [BP99, pp. 496-497], [HM04, pp. 214-220], and [KK09]. A declaration is the statement of an ID source on a specific attribute or behavior of the considered tracked object, based on emissions measured by sensors. Consider as example the declaration object follows a civil air route. We assume that each of N ID sources provides a declaration di ∈ Di with i ∈ {1, . . . , N}. An object belongs to exactly one of M possible Operational Object States. The finite set of all possible states is given by OOS := {O1, . . . , OM}, e.g., OOS = {OF,NA,EF} with own forces (OF ), neutral allegiance (NA), and enemy forces (EF ). For each ID source i the Source Likelihood Vector SLV i = (p(di|Oj))j=1,...,M is assigned to the declaration di. Fusion is performed by element-wise multiplication of the source likelihood vectors of all ID sources, providing the Combined Likelihood Vector CLV = (p(d1, . . . , dN |Oj))j=1,...,M , whereat conditional independence as a model-given precondition is assumed. Subsequently, by application of the Theorem of Bayes the Posterior Likelihood Vector PLV = (p(Oj |d1, . . . , dN ))j=1,...,M is calculated. Finally, the PLV needs to be transformed into a standard identity (Friend, Assumed Friend, etc.) to be presented to an operational user. A minimax approach can be used to select an appropriate identity based on the PLV , for details see [KK09]. By configuration, a Bayesian identification process can be customized for different operational and technical scenarios, see e.g., [KZ08]. Subsequently, we will define and analyze conflicts based on the sketched exemplary identification process. The treatment can be transferred easily to other Bayesian-based identification processes.