Hidden Markov chain (HMC) is the theoretical basis of traditional multi-target tracking. Based on the analysis of the limitations of HMC model, the more general pairwise Markov chain (PMC) model is introduced, and probability hypothesis density (PHD) based on PMC model filtering algorithm is deduced, and its Gauss mixture (GM) implementation is improved by using an elliptic gate to establish a reduced measurement set for each Gaussian component to update corresponding Gaussian component. The simulation experiment shows that the improved GM implementation of PMC-PHD filter achieves 1/3 of the original time without affecting the tracking precision in the scenario of high clutter density, and the simulation experiment also proves that the tracking performance of PMC-PHD filter for the adjacent targets in the HMC model scene is better than HMC-PHD filter.