Combining Model-based Reasoning with Knowledge Discovery Techniques for Level 2 and 3 Fusion

We propose to develop an approach to combine model based reasoning with knowledge discovery techniques for enhanced Level 2 and 3 data fusion, especially suitable for detecting asymmetric threats (e.g. ambush, insurgency) in cluttered urban environments. The knowledge discovery part: 1) deploys evidence filtering of large volumes of intelligence data to detect low-signature significant spatio-temporal events; and 2) uses clustering to perform spatial and time-series analysis of messages without requiring semantic information in the data. The former, for example, detects and tracks isolated suspicious vehicles, whereas the latter detects spatially correlated moving units over time within urban environments. Detected events and patterns trigger the need for assessing newly developed situations and threats, resulting in invocations of doctrine-based static and dynamic Bayesian belief network (BN) models that are causal and graphical in nature, and are well known for handling uncertainty. The selected BN models then perform higher-level data fusion based on other observables propagated as evidence into the models, by taking into account varying credibility and confidence of information sources via the Dempster-Shafer (D-S) theory of belief functions. The proposed hybrid approach will significantly enhance the fusion capability of DCGS-MC and C2PC for Marine Corps operations in urban environments.