Bayesian vs MAP Multiple Model Adaptive Estimation for Field of View Expansion in Tracking Airborne Targets

摘要

Previous efforts have led to the development of a multiple model adaptive filter (MMAF) tracking algorithm which demonstrates significant improvements in performance against close-range, highly dynamic, airborne targets, over a direct correlation method currently in use. The basic elemental filter in the MMAF bank combines an enhanced correlator and a linear Kalman filter. Digital signal processing techniques are used to derive a target shape function from the forward looking infrared sensor data. This shape function is used as a template in the correlation algorithm which generates offset pseudo-measurements for the update portion of a linear Kalman filter. The multiple models are created by tuning the basic model for best performance against differing target maneuvering behavior and with physically different fields of view. The outputs of three independent elemental filters, each receiving data from a shared sensor are used to generate a single adaptive estimate of the state via a probabilistic weighted average (Bayesian form) or by selection of the one elemental filter associated with the highest probability (MAP form). The adaptive state estimate can produce target position predictions to be used in generating feedback control for maintaining the target in the center of the field of view. There are two main results from this effort. The addition of a third elemental filter to the baseline MMAF improves tracking performance over the two-element MMAF. Specifically, the peak error following a maneuver is significantly reduced. However, the MAP estimation approach does not differ significantly from the Bayesian approach.