Fast data association approaches for multi-target tracking

Abstract

Gaussian-Mixture Probability Hypothesis Density (GM-PHD) filter is one of the implementation of PHD filter based on Random Finite Set (RFS). The algorithm performs well in jointly estimating the number of targets and their states with low computation demanding. However, the GM-PHD filter can't provide trajectories of individual targets. This paper proposes two approaches to combine the GM-PHD filter with the Multiple Hypothesis Tracking (MHT). On the one hand, GM-PHD filter effectively reduce the computation complexity of MHT; On the other hand, the data association problem is successfully solved by MHT. The simulation shows that the calculation cost is decreased remarkably and the association accuracy is improved at the same time compared with MHT.