Hybrid Inferencing for Data Fusion and Situation Assessment

We have developed a hybrid inference system for data fusion (DF) and situation assessment (SA) for multi-source intelligence analysis, with specific application to Marine Corps operations in urban environments. The approach views an urban environment as a complex dynamic system whose state vector is composed of a large number of both discrete and continuous variables (hence is hybrid) representing properties of tracked entities. Our inference algorithms exploit both causal dependencies among variables in the state vector via its Dynamic Belief Network (DBN) representation, and vector decompositions into weakly interacting subcomponents via factored Particle Filtering (PF). Under Phase I, the DF and SA functions were demonstrated and evaluated for an ambush scenario in Baghdad. For Phase II, we propose to develop a top-down, library-based methodology for systematically generating DBN models for SA. Incorporating richer dynamics into movement models, along with spatio-temporal clustering of units, will enhance the DF functionality for tracking units. The efficiency of PF algorithms will be enhanced via an asynchronous sampling mechanism. Improved tailoring and customization ability will be provided via libraries of domain expert-specified heuristics and plug-and-play inferencing algorithms. Evaluation will be based on well-defined performance metrics applied to a wide range of scenarios in urban environments.