Final Report on Contract F30602-91-C-0037 (Massachusetts University)
Abstract
Over the past twenty-plus years of computer vision research, a wide variety of algorithms have been developed to solve many visual subproblems, ranging from edge extraction to vanishing point analysis to geometric model matching. Despite these advances, however, very few systems have been built that exploit the information in images to solve practical problems. The problem is a lack of understanding of how these algorithms (and representations) should be combined; the goal of this contract was to investigate the use of machine learning techniques to automatically build executable object recognition systems out of these readily available components. This report describes a system, called the Schema Learning System (SLS). SLS does not try to match abstract object models directly to image data. Drawing on twenty years of computer vision research, SLS compares models to data by reasoning across multiple levels of representation. The computer vision literature contains many representational systems for visual data, as well as many algorithms for creating and evaluating instances of these representations. SLS integrates this research by selecting the visual procedures and representations that will best satisfy a particular goal, and building an executable control strategy for invoking those procedures to achieve the goal.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1991
- Accession Number
- ADA283511
Entities
People
- Allen R. Hanson
- Bruce A. Draper
- Edward M. Riseman
Organizations
- University of Massachusetts Amherst