Lie group parametrization for dynamics based prior in ATR
Anuj Srivastava, Michael I. Miller, Ulf Grenander · 2002
In recent automated target recognition (ATR) literature, there is increasing utilization of motion dynamics for tracking and recognizing moving targets. The dynamics along with the sensor models provide a Bayesian framework for conditional mean estimation of scene parameters. Previously, the authors have presented a random sampling algorithm for empirical generation of estimates based on jump-diffusion processes. Here we describe a different parameterization for simplifying the derivation of a more informative prior, from Newtonian mechanics, on the target configurations.>