Matching Pursuits & Hidden Markov Models for Processing IR Imagery

摘要

Infrared imagery from several military vehicles is considered, with the goal of classification. The IR imaging process effects a projection of a three-dimensional target onto a two-dimensional image. The target-sensor orientation is uncertain, as is the target identity. There are contiguous sets of orientations for which the imagery is statistically stationary, with these termed states of the target. Each target has multiple states, and these can be employed in a Hidden Markov model for moving-target classification. A set of expansion-matching (EXM) filters is constructed for the target parts, for a given target state. The output of the EXM filters are processed in a tree-like fashion, via a Hidden Markov Tree. Classification performance is assessed through consideration of multi-target IR imagery acquired from the US Army Night Vision and Electronic Systems Division (NVESD).