The Set MHT

Paper in proceeding, 2011

Abstract—We introduce the Set MHT, a tracking algorithm that maintains multiple hypotheses and produces “smooth” estimates without the track coalescence often associated with Minimum Mean Squared Error (MMSE) estimation or the jitter associated with Maximum Likelihood (ML) estimation. It does this by utilizing Minimum Mean Optimal Subpattern Assignment (MMOSPA) estimation techniques coupled with a theoretically-grounded approach for probabilistically determining the identities of the state estimates. Unlike traditional MHT algorithms, the Set MHT does not “forget” uncertainty in target identities, i.e. display an unjustifiably high confidence level in the target identities, as a result of pruning out competing hypotheses. Rather, it uses merging techniques while avoiding the shortcomings of traditional Gaussian mixture reduction trackers.