Estimation of object position using non-linear filters

Abstract

The paper presents chosen results of testing of non-linear filtering algorithms (an Extended Kalman Filter, two versions of Unscented Kalman Filters and a Particle Filter) in tracking applications. The accuracy of filters have been assessed and compared. The movement of tracked objects has been modeled in a Cartesian frame of reference, whereas the measurements are assumed to be realized in a polar frame of reference. The simulations have been realized under the assumption that the acceleration is described with the Univariate Non-Stationary Growth Model. All the tests have been performed in Matlab®.