Some Considerations in a Model Building System for Scene Analysis.
MASSACHUSETTS UNIV AMHERST DEPT OF COMPUTER AND INFORMATION SCIENCE
Pagination or Media Count:
This paper outlines the design of a system, called VISIONS, whose goal is to build a semantic 3-dimensional model from a 2-dimensional digitized scene. There are many kinds of information that must be employed in model construction, ranging from processed visual data to highly structured semantic information embodied in context frames. The modular subsystems that process this information interact through an executive which is responsible for the model construction. The authors discuss a variety of considerations in making such a system both flexible and feasible. Brief arguments are offered for dealing with confidences, expectations, and importance of objects, attributes, and partial models in a rough manner. A simplified scenario of model construction demonstrates how the system might work.