Ground Moving Target Tracking with VS-IMM Using Mean Shift Unscented Particle Filter

Abstract

In order to track ground moving target, a variable structure interacting multiple model (VS-IMM) using mean shift unscented particle filter (MS-UPF) is proposed in this paper. In model-conditioned filtering, sample particles obtained from the unscented particle filter are moved towards the maximal posterior density estimation of the target state through mean shift. On the basis of stop model in VS-IMM, hide model is proposed. Once the target is obscured by terrain, the prediction at prior time is used instead of the measurement at posterior time; in addition, the road model set used is not changed. A ground moving target indication (GMTI) radar is employed in three common simulation scenarios of ground target: entering or leaving a road, crossing a junction and no measurement. Two evaluation indexes, root mean square error (RMSE) and average normalized estimation error squared (ANEES), are used. The results indicate that when the road on which the target moving changes, the tracking accuracy is effectively improved in the proposed algorithm. Moreover, track interruption could be avoided if the target is moving too slowly or masked by terrain.