Random Shape and Reflectance Representations for 3D Assisted/Automated Target Recognition (ATR)
Abstract
This document is the final report for research on ATR Center RASER Grant FA8650-07-1-1113. The objective of this project was to expand the capabilities of model-based assisted/automated target recognition (ATR) systems by explicitly accommodating variation in shape and reflectance across elements of a broad target class. Work is set in the context of three-dimensional point-cloud data sets, such as LADAR or other structured light methods, and builds off a data representation model that represents measurement uncertainty probabilistically. Under this data model, the likelihood that a particular target gave rise to an observed point cloud can be computed using a collection of numerical integrations over the surface of a model of a target. Selection of the target with the largest likelihood then yields the classification result with the minimum probability of error (MPE) that can be achieved using a given sample of observed points. Our focus is on the study of anytime ATR algorithms, which are structured to support classification result queries that are placed at unknown, arbitrary times. A naive anytime algorithm based on the MPE decision rule can be defined in terms of round-robin calculations of likelihoods for observed points.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 2010
- Accession Number
- ADA516723
Entities
People
- B. M. Horowitz
- I. O. Reyes
- M. D. Devore
- Peter A. Beling
Organizations
- University of Virginia