A Generic Validation Methodology for Multispectral Synthetic Scene Generator Models - Interim Report
Abstract
The establishment of a sufficient, field-measured database to Support the analysis of automatic target recognition (ATR) algorithms, sensor fusion effectiveness, and sensor system performance for multiple combinations of targets, environments, sensors, and locations will severely challenge the limited, available resources currently within the U.S. Army research community. However, the use of a high-resolution, synthetic scene generator model (SSGM) for time-independent applications can alleviate the database requirement. We propose a methodology for a robust validation of SSGM that will consist of defining sets of images (real and corresponding SSGM imageries) and using human observers to define a baseline. First-order comparisons of a real scene to a synthetic scene will be performed with the use of the filters in the Tank-Automotive Research, Development and Engineering Center (TARDEC) model or a comparable computational vision model (CVM). The similarity of target-to-background histograms as a function of various CVM filters will need to be analyzed to define first-order effects. Second-order metrics are defined in terms of probability of detection, detection timeline, and false alarm rate. A metric for the target signature will be mathematically defined to test these second-order effects. For a given application, the necessary and sufficient metrics are discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1997
- Accession Number
- ADA332048
Entities
People
- Charles R. Kohler
- Marcos C. Sola
- Mark W. Orletsky
- Quochien B. Vuong
Organizations
- United States Army Research Laboratory