Autonomous decentralized multiplatform information fusion in littoral and riverine environments using dispersed and highly disparate heterogeneous sensors on unmanned systems is a major theoretical and practical challenge. Besides highly diverse information types, systems of this kind must deal with potentially large target numbers, closely-spaced targets, potentially dense clutter, limited communication bandwidth and intermittency. The Scientific Systems Company, Inc. (SSCI) team proposes a foundational approach, based on five innovations: (1) a multisensor-multitarget likelihood function f_{k+1}(Z_{k+1}|X) that encapsulates all relevant information regarding the characteristics of the various sensors situated on various platforms; (2) unified probabilistic representation and Bayesian processing of heterogeneous information types, such as radar, EO/IR images, acoustics, and even inference rules and natural-language statements; (3) a dynamic "tactical importance function" (TIF) that mathematically specifies the meaning of target prioritization ("tactical significance") for a given theater at any given moment, thus providing a statistical basis for automatic operator alerting; (4) integration of these concepts with track-before-detect filters; and (5)theoretically rigorous incorporation of the constraints due to the platform, terrain, and other communication-systems topologies and constraints. Under this approach, information from disparate fixed or mobile netted sensors---including those providing feature information---can be adaptively and optimally fused to create a common operational picture, based on a dynamically changing definition of target importance. Our project team includes Lockheed Martin, iRobot, and Kairos Autonomi. Lockheed Martin will provide both technical and commercialization support in the application of data fusion for Distributed Sensor Systems. iRobot and Kairos Autonomi will support fabrication of a prototype system in Phase II.
Keywords: Target Id, Multi-Target Tracking, Data Fusion, Track-Before-Detect Filters,
