Improved Multi-Bernoulli Filter for Extended Stealth Targets Tracking Based on Sub-Random Matrices

In this paper, we address a radar detection and tracking of multi-stealth targets. Multi-Bernoulli filter is a provable optimal Bayes approach of multi-target filtering and shows excellent performance over the other filters. The original multi-Bernoulli filters rely on the assumption that the interest targets are point source objects. Normally, in realistic scenarios, the point source assumption is not suitable for estimating the extent stealth targets (ESTs) scenarios. Recently, the random matrices approach has been proposed for ellipsoidal extended object tracking by additional state variables. In multiple EST scenarios, although the extension ellipsoid is efficient, it may not be accurate enough because of lacking useful information, such as size, shape, and orientation. To this point, we introduce a non-ellipsoidal EST composed of multiple ellipsoidal sub-objects, and each one is represented by a random matrix. Based on such models, a multi-Bernoulli filter is used to estimate kinematic states and extensions of sub-objects for each EST. However, in multiple EST scenarios, the detection profile is prior unknown due to fluctuation of EST parameters. The parameter, such as detection probability, is of critical importance in this extended filter. To tackle these problems, a beta Gaussian inverse Wishart implementation is proposed to estimate the EST extent augment with unknown detection probability and kinematic state. The results show that the proposed filter has more accurate cardinality estimation and smaller optimal sub-pattern assignment errors than the recently extended probability hypothesis density (PHD) and cardinalized PHD filters.