ABSTRACT
Tracking development has progressed slowly in terms of its ability to track under more stressing conditions. Rudolph Kalman’s1 innovative paper on linear ltering and prediction problems led the way to provide one of the rst adaptive and optimal approaches for realistic target tracking. As a result of the importance of this work, Kalman was awarded the Kyoto Prize ($340,000), considered to be the Japanese equivalent to the Nobel Prize.2 Since that time, the Kalman lter has become the cornerstone for most of the major tracking developments to date.3-7 Kalman lters represent the class of optimal Bayesian estimators that operate under conditions involving linear-Gaussian uncertainty. As long as the target dynamics operates within a polynomial motion model and sensor measurements are related linearly to the target state with Gaussian noise, the approach is an optimal estimator. Although the success of the Kalman lter led to numerous military and civilian applications, it has its problems8: to name a few, there is marginal stability of the numerical solution of the Riccati equation, small round-off errors accumulate and degrade
performance, and conversion of non-Cartesian sensor measurements introduces coupling biases within the Cartesian Kalman state. This did not stop researchers from developing methods to expand Kalman lter operating bounds to the larger nonlinear problem classes.
