Acquisition of Equipment for Research in Bayesian Automated Target Recognition
Abstract
The research reported here involves Bayesian ATR: development of statistical models and algorithms for statistical inference. The specific items that were covered are: (1) Statistical Models for Thermal Variation in Prediction of IR Images: We are interested in statistical tools for predicting IR images of a known target in a new, previously unobserved thermal state. This prediction is based on a partial observation of the new state and the database of previous observation. We have developed a linear regression framework for estimating the temperature field, associated with the new state, and using it for predicting IR images from arbitrary perspectives. (2) Statistical Models for Clutter: We have derived analytical forms, called Bessel forms, to model the marginal densities of the filtered images, filtered using a set of Gabor filters. These analytical forms are easy to compute and match well with the observed histograms. In addition, a closed-form expression for the L(sup A)2 metric, on the space of these densities, provides a measure of closeness between natural images, with applications in clutter classification. (3) Nonlinear filtering for Tracking of Manifold-Valued Parameters: In ATR and other signal/image processing applications, we are often interested in tracking parameters that are constrained to be manifold-valued. We have applied a sequential Monte Carlo algorithm to solve the nonlinear, non-Euclidean filtering problem on these manifolds. (4) Asymptotic Performance Analysis: Bayesian ATR corresponds to selection of hypothesis in the presence of nuisance variables. Using Laplace's approximation to integrate out the nuisance variables (pose, location, motion etc.), we have derived analytical forms for the probabilities of error in ATR. Additionally, we have quantified the relation between nuisance estimation errors and the ATR performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2001
- Accession Number
- ADA392698
Entities
People
- Anuj Srivastava
- Jayaram Sethuraman
Organizations
- Florida State University