Pattern Theoretic Bayesian Inference for Multisensor Fusion

The fusion of multiple sensor information is crucial to the near- and long-term development of improved automatic target detection and tracking. Recent advances allow analysis of practical sensor suites and more realistic scenarios. Our pattern theoretic approach unifies the traditionally separate endeavors of detection, tracking, and recognition. We postulate data likelihood models for sensors of interest. A posterior distribution is obtained by combining these with a track motion prior generated by stochastic differential equations. Conditional mean estimates for empirically generating estimates of target positions and types are generated using a random sampling algorithm based on continuous and discontinuous stochastic processes. New objects are detected and object types are recognized through discontinuous moves. The location of objects are estimated via continuous processes. The methodology outlined is universal and may be applied to any other sensor suite. Such an approach lays the basis for detection, understanding, and recognition of targets in the complex battle scenarios which will face the military in the future. We propose to implement these algorithms and analyses (theoretically and computationally) their ability to obtain and maintain lock under a variety of practical conditions, as well as their ability to handle the extreme conditions required in true use.