Comparison of MHT and PDA track initiation performance

The relative performance of multiple hypothesis tracking (MHT) and probabilistic data association (PDA) track initiation is investigated using simulated wide-area-surveillance radar data. The scenarios involve single targets that are low-observable (i.e. a low probability of target detection and a high probability of clutter detection) and non-manoeuvring. Divergent track counts and the receiver operating characteristics (ROCs) are determined and compared. The PDA implementation is extended to incorporate automatic track initiation using the concept of target visibility (PDAv). The MHT implementation uses a log likelihood track score to rank and prune hypotheses. A normalised overlap integral is used to eliminate duplicate hypotheses. The correlation between performance and the number of maximum number of retained MHT hypotheses was examined. A favourable linear performance/complexity relationship was observed. It was found that PDAv is more likely to suppress false tracks and to quickly confirm true tracks than MHT; however, the confirmed true PDAv tracks are more likely to be divergent.