Unified Generalized Bayesian Adaptive Data Fusion Technology

Real-time fusion algorithms are often patchworks of loosely integrated sub-algorithms, each of which addresses a separate fusion objective and each of which may process only one kind of evidence. Because these objectives are often in conflict, adaptive methods (e.g. internal monitoring and feedback control to dynamically reconfigure algorithms) are often necessary to ensure more optimal performance. The proposed project offers a plicit algorithm reconfiguration is largely unnecessary because conflicting objectives are simultaneously resolved within a self-reconfiguring, optimally integrated algorithm. This algorithm is ultimately capable of filtering all major evidence types: random point data, images, ambiguous features, natural-language statements, and rules. The approach, which we call Bayesian multisource, multi-target, multi-evidence filtering or Bayesian multi-filtering for short, is based on a direct generalization of recursive Bayes-Markov nonlinear filtering (WOLF) theory. It generalizes NLF to: (1) multiple targets of unknown number, identity, and kinematics, observed by multiple sensors; including (2) sensors which collect image data; and (3) sources which collect ambiguous observations. Specific Phase I tasks are: (1) Develop or acquire simulated multi-source data for multi-sensor sources; (2) Develop multi-source, multi-target, and multi-evidence unified evidence accrual algorithms; (3) Test and evaluate the novel evidence accrual algorithms; (4) Statistically characterize the performance error of the novel algorithms; and (5) Final report and Phase II recommendations. Phase II will further develop the chosen data fusion paradigm and evaluate it using appropriate metrics. The project team includes Dr. Ronald Mahler of Lockheed Martin and Dr. Anuj Srivastava, both originators of Bayesian multi-filtering algorithms. Lockheed Martin will provide technical and commercialization support in the application of Bayesian multi-filtering.