Event-averaged maximum likelihood estimation tracking for fire-control

The paper describes an event averaged maximum likelihood estimation (EAMLE) filter and compares its performance with that of a joint probabilistic data association (JPDA) filter in a low observable fire control application for tracking crossing targets in clutter. One of the main distinguishing features of the EAMLE filter is that it explicitly models the error correlation that arises between close targets. As the target separation goes to 0, their error correlation goes to 1, leading to a filter instability that must be regularized. With appropriate regularization, the EAMLE filter estimate has smaller mean square error than the JPDA estimate and lower track loss rate. For a 6 dB test problem studied here, JPDA looses about 5% of targets when they cross. Tracking is improved with EAMLE so that only about 2% of targets are lost.