Frederick Garber (Committee Member), William Pierson (Committee Member), Brian Rigling (Advisor)
Master of Science in Engineering (MSEgr)
We consider the problem of LADAR ATR classifier performance prediction in the presence of arbitrary nuisance parameters including but not limited to pose. We use several noise models for both range images and point clouds that are significantly more accurate and complex than the Gaussian models used by previous non-Monte Carlo prediction methods. Two accurate new methods of efficiently predicting the optimum Bayesian classification performance are then derived, and applied to the noise models. Advantages of these methods include significant gains in accuracy for medium to high noise levels and the ability to handle target near symmetry. Extensions are developed for multiple targets and predicting the performance of classifiers designed using incorrect noise models. We also derive several simple analytic approximations for the behavior of the probability of error as important sensor and noise parameters vary. Finally, we verify the accuracy of our predictions using Monte Carlo simulations.
Department or Program
Department of Electrical Engineering
Year Degree Awarded
Copyright 2012, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.