Alternative Dynamics Models and Multiple Model Filtering for a Short Range Tracker.

摘要

The performance of three extended Kalman filter implementations that estimate target position, velocity, and acceleration states for a laser weapon system are compared using various target acceleration trajectories. Measurements available to the extended Kalman filters each update are taken directly from the outputs of a forward looking infrared (FLIR) sensor. Two dynamics models considered for incorporation into the filter are (1) a Brownian motion (BM) acceleration model and (2) a constant turn rate (CTR) target dynamics model. The CTR filter was compared against the BM filter to see if the more complex dynamics of the CTR filter gave it a significant improvement in tracking performance over the BM filter. These two simple extended Kalman filters were then compared to a multiple model adaptive filter consisting of a bank of three filters based on the Brownian motion acceleration model. All three filters are tested using three different flight trajectory simulations: a 2 g, a 10 g and a 20 g pull-up maneuver. All evaluations are accomplished using Monte Carlo simulation techniques. The constant turn rate extended Kalman filter was found to outperform the other two filters. The main advantage this filter had was the minimization of mean bias error in estimating position. The standard deviation of error was also slightly lower in most instances. (Author)