Computational and Information Sciences Directorate Research Areas

Statistical Techniques of Data Fusion

Advisor: BA Bodt (barry.a.bodt.civ@mail.mil)
Aberdeen Proving Ground, Maryland

Key words: tracking (positions), targets, statistics, nonparametric statistics

Research is needed to expand mathematical/statistical bases for fusing battle space information. Fusion is desired at various levels. At the lowest level, multisensor acquisition systems provide information to allow for the detection and identification of battle space entities. Target detection, identification, and tracking must be approached in a systematic and theoretically sound fashion characterized by informative data representation, coherent pooling of diverse data, and information quality measurement (e.g., via likelihood, belief, or entropy).

Classical approaches might use logistic regression, discriminant analysis or decision trees. Scalable methods for representing and fusing multidimensional information from multiple sensors are sought.  At higher levels, context is most important. What is the intent and threat of the entities uncovered? In this much harder problem, innovative techniques are needed.

Many approaches have been proposed, including logical templating and Bayesian belief networks. Research that leads to development of goodness metrics and improved information processing under strong constraints on power and bandwidth is of interest. We invite proposals on emerging techniques that could be applied to data characterized by multivariate time-dependent structures.