A probability hypothesis density filter for tracking non‐rigid extended targets using spatiotemporal Gaussian process model

Abstract

Abstract This paper proposes a random finite set (RFS)‐based algorithm to deal with the tracking problem of multiple non‐rigid extended targets (MNRET) with irregular shapes in the presence of clutter, false alarms and missed detection. The extensions of targets are modelled by spatiotemporal Gaussian process, which is augmented with internal reference point (IRP) modelling the kinematics to construct the state of MNRET. The probability hypothesis density (PHD) filter is employed to propagate the first‐order moment of the RFS of MNRET. A suitable predicted likelihood of MNRET for the optimal partition is given, and the filter recursion is presented along with the necessary approximations and assumptions. More importantly, the closed‐form implementation and its corresponding smoothing filter are derived by converting the posterior density to Gaussian mixture form. Simulation results show the robustness and effectiveness of the proposed algorithm.

Keywords

• Gaussian mixture
• non‐rigid extended targets
• predicted likelihood
• probability hypothesis density
• random finite set
• Spatiotemporal Gaussian process