Unified Bayesian Cluster Target Tracking and Discrimination

Cluster tracking and discrimination is a major problem in ballistic missile defense.Here, one must track clusters, determine if they contain Reentry Vehicles (RVs), and locate and trackthe RVs. This problem presents a major theoretical and practical challenge becausethe assumptions underlying the currently most promising approach, Multi-HypothesisCorrelation (MHC), limit its effectiveness. Scientific Systems Co., Inc. (SSCI) andits subcontractor Lockheed Martin Tactical Systems (LMTS) believe, nevertheless,that a theoretically rigorous approach to cluster target tracking and discriminationis now feasible. Our approach is a direct generalization of Bayes-optimal recursivenonlinear filtering to the multisource-multitarget realm, made possible byfinite-set statistics (FISST). FISST provides a systematic means of dealing withthose applications, such as cluster tracking and discrimination, in whichthe observations and/or the states are randomly varying finite sets. FISST treats clusters as single unitary entities that differ from ordinarypoint targets only in that they generate far more complex sensor returns.True likelihood functions and true Markov motion-model densities formultitarget problems can be defined and constructed for clusters using aFISST generalization of the differential calculus. This includes, inparticular, multitarget motion models in which numbers of targets canchange, or in which target motions are correlated. The project team includes Dr. RonaldMahler of Lockheed Martin, who will provide both technical and commercializationsupport in the application of Cluster Target Tracking technologies.Multitarget detection, trackingand identification based on diverse evidence types is one of the keytechnologies for global surveillance, precision strike, air superiority anddefense, which are three of the seven science and technology thrust areasidentified by the Director of Defense Research and Engineering. The currentlimitations are due to poor understanding of how to model, fuse, and filterboth conventional and unconventional forms of evidence. The proposed R&Daddresses all of these problems.