A PHD filter for tracking closely spaced objects with elliptic Random Hypersurface models

It is one of the difficulties for space-based optical sensors to track a group of closely spaced objects (CSOs). A suitable way is to treat the CSOs as extended targets, tracking the centroid states and extensions jointly instead of individual objects. The target-generated measurements are modeled by the elliptic Random Hypersurface model (RHM) which can be easily embedded into other Bayesian inference algorithms. Here, the RHM is incorporated with the Gaussian-mixture probability hypothesis density (GM-PHD) for extended target tracking, which enables estimating shape-varying targets in presence of clutter and uncertain associations. Compared with the Gaussian inverse Wishart (GIW) PHD, simulation results demonstrate the capabilities and limitations of the proposed method. At last, a ballistic missile defense scene is established to depict how the method is to handle the CSOs tracking issue.