Combined Kalman and Kalman-Levy filter for maneuvering target tracking

Common target tracking algorithms, such as the Kalman Filter, assume Gaussian estimates of process and measurement noises. This Gaussian assumption does not fully support practical maneuvering target tracking. Rather, when target motion is highly dynamic, sudden maneuvers are better described by non-Gaussian noise distributions. A Kalman-Levy filter has been proposed as an improvement to the maneuvering target tracking problem. This filter models process and measurement noises using Levy distributions. While an improvement in maneuver estimation is demonstrated with the Kalman-Levy filter, it requires significant computation time and occasionally provides poor estimates of simple, linear maneuvers that the Kalman filter can otherwise provide. This paper seeks to improve maneuvering target tracking without sacrificing computation time by proposing the use of a moving-average filter in the tracking process. A Moving-Average filter is used to track the position root-mean-square error (RMSE) and switch from the Kalman filter to the Kalman-Levy filter when this error becomes large. The Kalman filter, the Kalman-Levy filter, and the switching algorithm based on the Moving-Average filter are demonstrated on two tracking problems. Simulation results show that switching between the filters improves maneuvering target state estimation accuracy while being computationally efficient.