Advanced Automatic Target Recognition.
Abstract
This final report summarizes the findings of the research, Advanced Automatic Target Recognition, supported by AFOSR grant P4962O-97-I-O523. In this research effort, we have developed a method for detecting buildings from SAR images, so that false alarms due to building returns can be reduced. We consider three scenarios corresponding to incremental data availability from a high-resolution, airborne SAR, multiple SAR images and interferometric SAR. In a single strip-map SAR image, we look for certain characteristics exhibited by buildings in radar imagery, namely the combination of cardinal streaks and supporting shadow, to delineate buildings. We then present a framework for registering multi-pass airborne SAR images and for extracting heights of 3-D structures which produce identifiable linear patterns in them. Finally, given noisy elevation data derived from an interferometric (IF) SAR, buildings are segmented from the ground using a local histogram-based thresholding scheme, consolidated by propagating the thresholds, and refining along their edges to reduce errors. The effectiveness of the building detection and height estimation algorithms is demonstrated using examples of high-resolution SAR data from Lincoln Laboratory's ADTS radar and elevation data derived from Sandia's IPSAR platform. Our results will make possible on-the-fly, context-based exploitation of SAR images.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 15, 1998
- Accession Number
- ADA341102
Entities
People
- Ram Chellappa
Organizations
- University of Maryland