A GM-PHD Filter for New Appearing Targets Tracking

Simulations reveal that the usual implementations of the Gaussian Mixture PHD filter can detect new targets only if its target-birth model is based on a priori knowledge of where new targets might appear. Otherwise, it cannot detect new targets (unless they happen to be near existing tracks) since it prunes Gaussian components that are not associated with existing tracks. In this paper, this problem is remedied by reserving at least one Gaussian component corresponding to each measurement in the revised Gaussian components pruning approach. Simulations involving four targets show that the proposed approach successfully deals with newly appearing targets.