Modified Scoring in Multiple-Hypothesis Tracking.
Stefano P. Coraluppi, Craig A. Carthel · 2012
Track-oriented multiple-hypothesis tracking is a powerful and widely-accepted methodology in multi-target tracking. We show that the target-death problem inherent in the probability hypothesis densityfilterdoesnotariseintheMHT.However,theMHTsuffers from a problem of its own: excessive competition for measurements from tentative tracks. We introduce a mechanism to mitigate this effect by favoring confirmed tracks in the association process. A heuristic justification for the technique is that it mitigates the suboptimality associated with hypothesis pruning and sequential track extraction. Perhaps more convincingly, the modification to the MHT equations is provably optimal in the limiting case of cardinality tracking with unity detection probability. We show that modifiedscoring MHT improves upon standard MHT in several benchmark studies.