Incorporating amplitude information into the PMHT using shot-noise models

Summary form only given. In the past several years the probabilistic multi-hypothesis tracking (PMHT) algorithm has been extended by several authors in a variety of directions. In this paper the PMHT is extended to directly incorporate amplitude by using marked shot-noise (or, compound point process) models to represent the spatial and spectral distributions for stochastic target signals and noise. Incorporating amplitude information into the multitarget tracking algorithm has several potential advantages. This extension may allow the PMHT to process sensor data directly, eliminating the need for a separate detection preprocessing subsystem. This extension also may eliminate the need for a separate background normalization preprocessor because it incorporates the spatial and spectral distribution of the noise into the PMHT likelihood function. Direct use of the amplitude information may yield more accurate tracks, particularly at low signal to noise ratios. In addition, the marked shot-noise model allows one to easily calculate explicit Cramer-Rao lower bounds (CRLBs) for multi-target tracks in clutter. The CRLB on the location of a Gaussian target in uniform clutter is presented as a simple example. The spatial and spectral models of targets and noise are used to parameterize the likelihood function of the sensor data. As in PMHT, the expectation-maximization (EM) method is used to derive maximum likelihood estimates of the target parameters. Marked shot-noise models are used to interpret the spatial and spectral models of the targets and the noise to derive the complete data likelihood function for the EM method