IMMJPDA versus MHT and Kalman filter with NN correlation: performance comparison

In a tracking problem a radar periodically scans the volume under surveillance and provides detections (plots) that indicate a target presence. Multitarget tracking systems in operational use today generally adopt Kalman filter (KF) techniques (coupled with a manoeuvre detector to introduce some kind of adaptivity), and nearest neighbour (NN) correlation. Today there are two new approaches to the tracking problem, namely: interacting multiple model joint probabilistic data association (IMMJPDA) and multiple hypothesis tracking (MHT) which promise improved tracking performance. The paper provides a performance comparison between these three tracking algorithms in terms of track maintenance probability and tracking errors. The NN + KF algorithm is used as reference because of its widespread use. Results show that MHT is superior to IMMJPDA and, as expected, both perform better than NN + KF; the cost of additional performance is increased, yet feasible, computing power