Representation, Modeling and Recognition of Outdoor Scenes
Abstract
The primary goal of this project was to advance the state-of-the-art in scene interpretation for autonomous systems that operate in natural terrain. In particular, techniques are being developed for representing knowledge about complex cultural and natural environments so that a computer vision system can successfully plan, navigate, recognize and manipulate objects and answer questions or make decisions relevant to this knowledge. The results include the development of representations and associated methods for rapidly modeling natural terrain, from multiple images, at a level or organization higher than that of the conventional dense array of depths. In particular, we have developed an approach for integration of information acquired from multiple views of a scene that uses a new class of geometric primitives; it allows easy expression of known constraints and observed data, and also allows the use of practical optimization based solution techniques. This work will provide an effective way of allowing a robotic system to incrementally build a progressively more accurate and complete model of the environment in which it is operating. Work is also progressing on the problem of recognizing import classes of natural and manmade objects -- especially roads, trees, rocks, and terrain features.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1994
- Accession Number
- ADA283361
Entities
People
- Martin A. Fischler
- Robert C. Bolles
Organizations
- SRI International