Experimental results on Bayesian MHT for maneuvering closely-spaced objects in a densely cluttered environment

Data association and resolution conflicts are inherent problems for tracking in a dense target/dense clutter environment, a fact taken into account by the multiple hypothesis character of many modern approaches. The task is made even more difficult by maneuvering targets, low scan rates, and imperfect detection. We discuss experimental results from real data (long-range radar) with dog-fight scenarios under serious operational conditions. Particular emphasis is placed on two items: 1) by explicitly handling of resolution conflicts in combination with the association task, the filter performance can substantially be improved. 2) Retrodiction, a generalization of standard smoothing to multiple hypothesis tracking (MHT), provides unique and accurate trajectories from ambiguous filtering output. Even a small delay can significantly ease the surveillance mission.