An EM-IMM Method for Simultaneous Registration and Fusion of Multiple Radars and ESM Sensors

Abstract

AbstractIn a multiple sensor tracking scenario, measurements originated from the target of interest are necessarily aligned and fused to provide accurate information about the target. The process known as registration and fusion is generally casted as a joint parameter and state estimation problem for a single target case. The standard solution to this problem is the augmented Kalman filter (AKF) which takes the parameters as variables in the state vector. Despite of its easy implementation, the AKF is not favorable for large number of unknown parameters, as is the case for multiple sensors. Moreover, it is prone to numerical inaccuracy or divergence in application. In this paper, we evaluate the divergence problem of the AKF in simultaneous registration and fusion for the Radar/ESM sensors. Furthermore, we propose an expectation-maximization (EM) method in the maximum likelihood estimation (MLE) framework. In particular, to account for the maneuverability of the target, the interacting multiple model (IMM) filter implemented either by an extended Kalman filter (EKF) or by an unscented Kalman filter (UKF) is embedded into the conditional expectation evaluation in the E-step. The proposed joint registration and fusion method is thus called EM-IMM. Analysis shows that the EM method is convergent and furthermore leads to asymptotically unbiased estimate in an approximation sense. To evaluate the estimation performance, a direct inverse computation algorithm of Fisher information matrix (FIM) in posterior Cramer-Rao bound (PCRB) is also developed. Simulation results are given to demonstrate the effectiveness of the proposed method.KeywordsExtended Kalman FilterFisher Information MatrixUnscented Kalman FilterSensor BiasSimultaneous RegistrationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.