A New Recursive Filter Based on the Gauss Von Mises Distribution

Many state estimation or sensor fusion algorithms are based on traditional filtering techniques such as the Kalman filter, the extended Kalman filter (EKF) or the unscented Kalman filter(UKF). However, these approaches all make Gaussian assumptions, without taking into account the intrinsic structure of the underlying state space. In this paper, to properly perform estimations on a cylindrical manifold, Gauss von Mises(GVM) distribution model is employed, a new GVM parameter estimation method is presented, and a novel GVM filter(GVMF) is proposed. The effectiveness of the proposed GVMF is illustrated in a case study in motion estimation involving the tracking of an object in a three-dimensional state space. Results demonstrate that more accurate estimates can be achieved with the proposed GVMF in comparison to the traditional EKF.