Fusion of Kinematic and Identification (ID) Information

ABSTRACT: Current approaches for advanced multi-sensor feature-aided kinematic tracking, including distributed multiple-hypothesis tracking (MHT), cannot contend with informative but sparse information that arises from some classes of physical sensors as well as from human input to the fusion process. The difficulty stems from the fact that, to deal with this multi-INT fusion challenge, computationally intractable hypothesis tree depths are required, even with efficient distributed MHT solutions.Graph-based tracking (GBT) methods have shown significant promise in providing scalable multi-sensor tracking solutions. Unfortunately, existing methodologies are not directly applicable to the multi-INT fusion challenge.Under this effort, STR will leverage and extend MHT and GBT technology to address the multi-INT fusion challenge. Additionally, we will introduce solution refinement and uncertainty representations via advanced machine learning technology. The refinements are based on efficient track-level Markov Chain Monte Carlo (MCMC) methods.The STR multi-INT data association solver (MIDAS) will exploit and advance the complementary nature of MHT, GBT, and MCMC solution paradigms, yielding an effective and scalable tool for large-scale forensic surveillance. Testing will be based on simulated scenarios and several multi-INT scenarios of current interest to AFRL.; BENEFIT: Under this effort, STR will develop the multi-INT data association solver (MIDAS) that will provide an innovative fusion capability that significantly extends the state-of-the art in MHT-based, graph-based, and machine-learning algorithms. MIDAS will enable fusion over large temporal and spatial datasets, over a wide range of sensor types with highly disparate data flows.