Improved Gaussian Mixture PHD Smoother for Multi-target Tracking

The Gaussian mixture probability hypothesis density (GM-PHD) smoother proposed recently can yield better state estimates than the GM-PHD filter. However, there are two major problems with it. First, the smoothed PHD distribution can not provide a more accurate target number estimate due to the target number estimation bias becoming larger by smoothing. Second, the computational complexity of computing the smoothed PHD distribution increases with the cardinality of measurement set, which can be very time-consuming when the clutter rate is high. To solve these problems an improved GM-PHD smoother is proposed that improves the target number estimation performance by using the estimated target number of forward GM-PHD filter and reduces the computational cost of GM-PHD smoother by the rectangular gating method. Simulated results show that the improved GM-PHD smoother is superior to the GM-PHD smoother in both the aspects of target number estimate and computational cost, so this improved GM-PHD smoother will have an applicable potential in related fields. Key-Words: Gaussian Mixture, Probability Hypothesis Density, Filtering, Smoothing, Target Tracking, Random Finite Set, Sequential Monte Carlo