Font Size: a A A

Vision in context: Using special purpose routines to achieve more general purpose vision

Posted on:2004-10-16Degree:Ph.DType:Thesis
University:Northwestern UniversityCandidate:Flachsbart, Joshua DavidFull Text:PDF
GTID:2466390011974673Subject:Computer Science
Abstract/Summary:
Recently researchers have used task driven computer vision routines to solve specific problems. These routines have been successful by relying on specific traits of the task they solve and the environment in which they operate. One disadvantage of this method is that the specific constraints on which the routines rely may not hold in a given situation, or may even cease to hold in situations where they held before due to changes in the world. We believe that reasoning about those constraints explicitly and selecting appropriate special purpose routines for specific situations allows us to create a general purpose computer vision system both robust and broad in task coverage.; This dissertation presents our work testing this hypothesis. The Intelligent Classroom is our primary testbed for utilizing vision in different contexts. The Classroom is an automated lecture facility that acts as its own audio/visual technician. The Classroom environment is a good choice for this research because it provides interesting computer vision tasks by interacting with a human user, and it also has structure which provides constraints for reasoning about that computer vision.; In addition, we present a framework for low level computer vision, and two additional systems which utilize that framework to solve similar problems in different domains.; The main contributions of the work described here are: (1) The Intelligent Classroom, a system which is able to use computer vision to produce a video of a lecture, as well as reason about how to sense for different user interaction tasks. (2) Our vision system, Gargoyle, which provides a framework for constructing vision routines on the fly, which is required to allow reasoning systems to manage the vision routines. It also provides a modular interface for designing and reusing parts of vision routines. (3) An implementation of existing vision routines in a robotic system, CHIP, utilizing this framework. This demonstrates the capability of the framework to implement proven systems. (4) The Interactive Image Mosaic, a system which demonstrates the ease of reuse by being built quickly using a number of the Classroom's Gargoyle routines.
Keywords/Search Tags:Routines, Vision, Purpose, System, Specific, Classroom
Related items