Improved measurement-driven Gaussian mixture probability hypothesis density filter

The probability hypothesis density (PHD) is an effective method for tracking the time-varying number of targets in multi-target tracking. Gaussian mixture is an approximation method to obtain the closed solution of PHD. However, the tracking performance of the Gaussian mixture PHD filter will decline sharply when multiple targets born and disappear in closely spaced target tracking scenarios. In addition, real-time performance of multi-target tracking cannot be met in heavy clutter scenario. To solve these problems, an improved measurement-driven Gaussian mixture PHD algorithm is proposed in this paper. First, the multi-target measurement set at each time step is divided into non-intersect measurement subset, where only survival and birth measurement set are used to update targets. Due to most clutter measurements do not used to update targets in the update step, better real-time performance can be achieved. Second, for the purposed of further improve the performance or multiple target tracking, a backward smoothing based on varied length window is utilized to reduce the possibility of wrong tracking of targets. In numerical experiments, the results demonstrate that the proposed approach can achieve better performance compared to the other existing methods.