A Performance Comparison of the PMD and IMM filters for a Mix of two Distinctively Different Classes of Target Trajectories

The purpose of the paper is a performance demonstration of the recently introduced filter, the probability mass diffusion (PMD) filter. The application is radar tracking of targets. The estimators are optimized for each of two distinctively different classes of targets and for a 50/50% mix of the classes. The estimators used are the PMD, IMM and the Kalman filter. Both the PMD and IMM are run with two and four models, respectively. The two classes of target trajectories are agile, highly maneuverable military type aircraft and slowly maneuvering, airline type aircraft, respectively. The evaluations are performed by the use of Monte-Carlo simulations, where the target trajectories are randomly selected from distributions describing the two classes of targets. The paper shows that the performance of the PMD filter is equal or better than the IMM for the cases studied. The improvements are shown as better noise-suppression during non-maneuver segments with comparable maneuver performance. The paper also shows the importance of trajectory definitions and filter tuning. Performance advantages may differ considerably depending on the defining set of target trajectories and the priorities of the user