Pattern-theoretic automatic target recognition for infrared and laser radar data

Pattern theory, a mathematical framework for representing knowledge of complex patterns developed by applied mathematician Ulf Grenander, has been shown to have potential uses in automatic target recognition (ATR). Prior research performed in the mid-1990s at Washington University in St. Louis resulted in ATR algorithms based on concepts in pattern theory for forward-looking infrared (FLIR) and laser radar (LADAR) imagery, but additional work was needed to create algorithms that could be implemented in real ATR systems. This was due to performance barriers and a lack of calibration between target models and real data. This work addresses some of these issues by exploring techniques that can be used to create practical pattern-theoretic ATR algorithms. This dissertation starts by reviewing the previous pattern-theoretic ATR research described above and discussing new results involving the unification of two previously separate outcomes of that research: multi-target detection/recognition and thermal state estimation in FLIR imagery. To improve the overall utility of pattern-theoretic ATR, the following areas are re-examined: 1) generalized diffusion processes to update target pose estimates and 2) the calibration of thermal models with FLIR target data. The final section of this dissertation analyzes the fundamental accuracy limits of target pose estimation under different sensor conditions, independent of the target detection/recognition algorithm employed. The Cramér-Rao lower bound (CRLB) is used to determine these accuracy limits.Ph.D