Comparison of Stochastic Integration Filter with the Unscented Kalman Filter for Maneuvering Targets

Sigma-Point Filtering (SPF) has become popular to increase the accuracy in estimation of tracking parameters such as the mean and variance. A recent development in SPF is the stochastic integration filter (SIF) which has shown to increase estimation over the Extended Kalman Filter (EKF) and the Unscented Kalman filter (UKF); however, we want to explore the notion of the SIF versus the UKF for maneuvering targets. In this paper, we compare the SIF method with that of the KF, EKF, and UKF, using the Average Normalized Estimation Error Square (ANEES) for non-linear, non-Gaussian tracking. When the nonlinear turn-rate model is similar to the linear constant velocity model, all methods are the same. When the turn-rate model differs from the constant-velocity model, our results show that the UKF with a large number of sigma-points performs better than the SIF.