Image Understanding at the University of Rochester

Christopher M. Brown and Randal C. Nelson
Department of Computer Science
University of Rochester
Rochester, NY 14627

Abstract: The vision group at Rochester is engaged in investigating several aspects of parallel and real-time computer vision with the overall goal of implementing a set of basic sensory-motor behaviors which could serve as a foundation for more sophisticated abilities, and integrating these primary behaviors into multi-modal systems. The emphasis is on behaviors which have relevance to, and can be implemented to work robustly in, a broad range of real-world environments since these are most likely to be useful as fundamental skills.

Our recent work includes commissioning the Rochester Robot, a 3 degree of freedom, two-eyed robot head mounted on a Puma 761 arm, and connected to a Datacube image processor. Several real-time visual behaviors have been implemented, including a vestibulo-ocular reflex (VOR), vergence, and target tracking. Research was also performed in various theoretical aspects of computer vision including parallel evidence combination, parallel object recognition, principal view analysis, and extended Kalman filtering.