Comparison of implementations of Gaussian mixture PHD filters

The Probability Hypothesis Filter, which propagates the first moment, or intensity function, of a point process has become more and more popular to address multi-tracking problems. Under linear-Gaussian assumptions, the intensity function takes the form of a mixture of Gaussian kernels. As the number of elements increases exponentially over time, deterministic pruning and merging steps are commonly used to keep the complexity bounded. In this paper, we study alternative stochastic strategies. The different strategies are compared on different scenarios. A new pruning strategy that maintains confirmed targets is also proposed.