Improved GMPHD tracker with delayed decision

The Gaussian mixture probability hypothesis density (GMPHD) filter was proposed to estimate the time-varying number of targets and their states, and GMPHD tracker even gives the identities of individual target state estimates, by recording the evolution of each Gaussian mixture component based GMPHD filter. However, in the presence of high density clutter and severe miss-detection, GMPHD tracker in the existing literature still has more or less defects, such as providing discontinuous track, or unsmooth track. In this paper, we propose and realize an improved GMPHD tracker with delayed decision, which implements a novel pruning and estimating scheme reducing the computational load and simultaneously provides more accurate target numbers and state estimates by a back scheme.