GENERAL METROPOLIS-HASTINGS JUMP DIFFUSIONS FOR AUTOMATIC TARGET RECOGNITION IN INFRARED SCENES

To locate and recognize ground-based targets in forward-looking IR (FL IR) images, 3-D faceted models with associated pose parameters are for mulated to accommodate the variability found in FLIR imagery. Taking a Bayesian approach, scenes are simulated from the emissive characteris tics of the CAD models and compared with the collected data by a likel ihood function based on sensor statistics. This likelihood is combined with a prior distribution defined over the set of possible scenes to form a posterior distribution. To accommodate scenes with variable num bers of targets, the posterior distribution is defined over parameter vectors of varying dimension. An inference algorithm based on Metropol is-Hastings jump-diffusion processes empirically samples from the post erior distribution, generating configurations of templates and transfo rmations that match the collected sensor data with high probability. T he jumps accommodate the addition and deletion of targets and the esti mation of target identities; diffusions refine the hypotheses by drift ing along the gradient of the posterior distribution with respect to t he orientation and position parameters. Previous results on jumps stra tegies analogous to the Metropolis acceptance/rejection algorithm, wit h proposals drawn from the prior and accepted based on the likelihood, are extended to encompass general Metropolis-Hastings proposal densit ies. In particular, the algorithm proposes moves by drawing from the p osterior distribution over computationally tractible subsets of the pa rameter space. The algorithm is illustrated by an implementation on a Silicon Graphics Onyx/Reality Engine. (C) 1997 Society of Photo-Optica l Instrumentation Engineers.