In multi-target tracking problem, not only the states of targets, the time varying number of targets, but also a sequence of observation sets in the presence of data association uncertainty, detection uncertainty, noise and false alarms should be estimated. The Gaussian mixture probability hypothesis density (GMPHD) filter offered an effective method for multi-target tracking. Due to the PHD propagation equations involved multiple integrals; there were no computationally tractable closed form expressions. Fortunately, the GMP HD filter provided a closed form solution to the PHD filter recursion. The posterior intensity function was estimated by a sum of weighted Gaussian components whose means, weights and covariances can be propagated analytically in time. Experiments show that the GMPHD filter can track a changing number of targets robustly, achieving near-real-time performance.