Adaptive target birth intensity for Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter

Abstract

In standard formulation of Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter, the newborn target intensity function is regarded as a known prior probability. This assumption limited the application in practice. An improved method is proposed based on the standard GMPHD by introducing logicals to differentiate two types of targets, called UGM-PHD filter. In the prediction step, if the logicals is equal to one, newborn targets are created from the received measurements at each scan. While in another situation, the intensity function corresponding to the both types of targets are added together and predicted jointly as same as the prediction step of persistent targets in the traditional GMPHD. Then in the update step, only the updated intensity function of persistent targets is concerned, since the updated weight of new targets will not exceed the output threshold. In this way, the target birth intensity can be obtained adaptively. By comparing the improved method with the traditional GM-PHD method, the simulation results show that the former improves the ability of searching newborn targets and the estimation accuracy of the number of targets.