Evaluation of Measurement Converted KF, EKF, UKF, CKF, and PF in GMTI Filtering

Abstract

The ground moving target indicator (GMTI) radar plays an important role in surveillance, precision tracking, and classification of ground moving targets. The measurements of a GMTI radar are range, azimuth, and radial velocity, which are nonlinear functions of the Cartesian target and sensor states. In our 2003 SPIE paper we analyzed the performance of the converted measurement based Kalman filter (CMKF), extended Kalman filter (EKF), unscented Kalman filter (UKF), and transition density based particle filter (TD-PF). In this paper, we extend our previous work by considering the cubature Kalman filter (CKF). We also compute the posterior Cramér-Rao lower bound (PCRLB) to judge the accuracy of the filters relative to the best achievable accuracy. The results of this analysis are useful in an operational system where a filter has to chosen based on state estimation accuracy, scalability, and computational complexity. We also analyze the accuracy of the TD-PF by varying the number of particles. We perform Monte Carlo simulations to analyze the accuracy and computational complexity of various filters. Based on the root time-averaged mean square (RTAMS) error of position and velocity, we find that the CMKF, EKF, UKF, and CKF have nearly the same accuracy in cases analyzed. To achieve the same level of accuracy, the TD-PF requires more than 40,000 particles with about 508 times more computational cost relative to the EKF. Contrary to conventional thinking, our results show that for cases analyzed the EKF produces the best accuracy with stable performance and lowest computational cost.