Multi-hypothesis, kinematic trackers are the state-of-the-art in automated GMTI data processing. These systems are not designed to recognize long duration behaviors and under complex conditions these systems often produce track snippets. Our approach, which we call Cognitive Fusion because of it’s structural similarity to human analysis methods reframes the problem from one of tracking all entities all of the time to one of tracking only behaviors the user cares about. CFUS adds value to tracking and fusion under real-world conditions that include moderate contact densities, unpredictable target motion, deception, and unreliable sensor returns. CFUS applies an abductive reasoning approach that combines hypotheses projection, contextual reasoning, and dynamically constructed Hidden Markov Models (HMMs) to find and track instances of hypothesized behavior in GMTI data. We demonstrate, using simulated data, how our algorithms can be used to find complex behaviors in cluttered data sets and can significantly reduce association false positives.
Reference:
Crossman, J., et al., Top-Down Abduction for Behavior Detection in GMTI Data, in Proceedings of the 14th International Conference on Information Fusion (FUSION) 2011. 2011: Chicago, IL. p. 1-8.