Improved Best-fitting Gaussian Approximation PHD Filter

Abstract

The best-fitting Gaussian approximation Probability Hypothesis Density (PHD) filter is a novel algorithm for multiple maneuvering target tracking. However, there is a problem that the model probabilities are calculated without the measurement innovation. To solve this problem, an improved algorithm is proposed in this paper, which develops an update procedure for model probabilities to employ the posterior measurement innovation to enhance the filtering performance. Then, the dynamic equations can be softly switched among different models according to the likelihood functions. The simulation results show that the improved algorithm has the advantages over the ordinary one in the aspects of target number estimation and filtering accuracy, implying good application prospect.

Keywords

• Random finite sets
• Probability Hypothesis Density (PHD) filter
• Best-fitting Gaussian approximation
• Maneuvering target tracking