Ascender II - Knowledge-Directed Image Understanding for Site Reconstruction.
Technical rept. 30 Apr 97-31 May 98,
MASSACHUSETTS UNIV AMHERST DEPT OF COMPUTER SCIENCE
Pagination or Media Count:
This report summarizes research done during the first year of the University of Massachusetts DARPA APGD contract. Our long term goal is the development of functional systems for detecting, confirming, and modeling building and other cultural objects from aerial images. The Ascender 2 system is a knowledge based system that uses multiple image understanding algorithms and a Bayes network controller that automatically selects and sequences IU algorithms for aerial image analysis tasks. A significant component of Ascender 2 is a system designed to generate 3-D models of buildings from elevation maps. The system automatically segments a digital terrain elevation map into coherent surfaces and reconstructs a 3-D model of the scene. Finally, Ascender 2s range of accessible imagery has been extended to include IFSAR data. A system has been developed that uses knowledge of radar sensor performance at building edges to hypothesize fragments of building boundaries. This information is fused with higher level knowledge about the geometry of buildings, such as constraint on their size, shape, and height, to locate boundary fragments in the image. All systems described in this report have been tested on real imagery, including the Fort Benning MOUT site, Kirtland Air Force Base, Fort Hood, and various ISPRS datasets.
- Cartography and Aerial Photography