Reduced cubature Kalman filtering applied to target tracking

Abstract

In a recent paper, a new discrete-time Bayesian filter, named the cubature Kalman filter (CKF), was derived. To reduce the complexity of the filter, we propose in this paper to combine the CKF with the linear Kalman filter, when either the process equation or the measurement equation is linear. The resulting filter is referred to as the Reduced CKF (RCKF). It is here applied to the problem of tracking in Cartesian coordinates a moving object whose state can be modeled by a linear dynamic equation, but whose measurement equation is non linear, due to the fact that the measurements represent position measurements in polar coordinates. The simulations results show that, in terms of root Mean Square Error (RMSE), the RCKF and CKF have the same performance, but the processing time of the RCKF is lower than that of the CKF.