Multiple manoeuvring targets tracking via GM-PHD and IMM-SB/MHT
Shicang Zhang, Liangbin Wu, Xinmei Hu · 2013
Probability hypothesis density (PHD) filter, which propagates the PHD or the first order moment instead of the full multitarget posterior density, is a promising multi-target tracking algorithm under density clutter capable of tracking random spawn and birth targets based on random finite set (RFS) statistics. Gaussian Mixture probability hypothesis density (GM-PHD) filter derivated from PHD filter plays an important role for PHD to practical applications. The interacting multiple model (IMM) technique can not be used in PHD and its derivations because the densities in PHD are not necessarily Gaussian and might be multi-modal. This paper proposes a combining technique which integrates linear Gaussian jump Markov system GM-PHD (LGJMS-GMPHD), structured branching multiple hypothesis tracking (SB/MHT) and IMM approach for multiple maneuvering targets tracking. First, LGJMS-GM-PHD is acted as pre-filter which eliminates most of clutter in measurements. Then, a gating technique is used according to state output of the LGJMS-GM-PHD, measurement which locates in the gate is a candidate measurement for branch and update under the framework of SB/MHT. Last, IMM is used to estimate the state of the maneuvering target in the framework of SB/MHT. This combining approach takes use of the advantages of LGJMS-GM-PHD, SB/MHT and IMM which can perform multiple maneuvering targets tracking under density clutter effectively. (5 pages)