The GFMT HPMHT puzzle

Most target tracking algorithms work with data at the plot or detection level; that is, after an initial signal processing step of thresholding and centroiding that delivers point “hits” for data association and filtering. The GFMT (general frequency modulation tracker) and HPMHT (histogram probabilistic multi-hypothesis tracker), on the other hand, work directly with pixellated observation data, as on a focal plane array (FPA), meaning that the signal processing step is integral to the tracker. The HPMHT filled most of a statistical pit in the GFMT, but a perceptible hole remains. We discuss this here: If the problem can be modeled - and it can - then a solution that derives from this model ought to be available.