Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision


Pose Estimation and Recognition Using Visual Attention

Greg Dudek
Associate Professor
McGill University
School of Computer Science and Center for Intelligent Machines

currently visiting
Xerox PARC, DiVA
and
Stanford University Robotics Lab
Monday, Oct 16, 2000, 4:15PM
TCseq200, Lecture Hall A
http://robotics.stanford.edu/ba-colloquium/

Abstract

In this talk I will consider the use of attention-like mechanisms for robotics positioning, modelling and recognition tasks. In particular, I will describe recent results on the use of attention operators to produce an efficient visual pose estimator for a mobile robot to recognized familiar objects and to build maps. Traditional methods for these problems use geometric scene models. To avoid the need for a geometric scene model, I will then consider appearance-based methods. By using raw sensor images, from either sonar or video, we can accurately estimate the position of our robot with only a very coarse initial position estimate. This is accomplished by storing an ensemble of highly compressed images which capture the variations in appearance of the environment. This can be accomplished using either global images features or local image-domain landmarks. An attention operator inspired by model of biological perception is used to select candidate landmark regions which appearance variations are subsequently learned to produce a complete pose estimator. The presentation is based on ongoing research with Robert Sim, Deeptiman Jugessur and Eric Bourque.

About the Speaker

Gregory Dudek is a Visiting Associate Professor at Stanford and an Associate Professor with the School of Computer Science at McGill University and a member of the McGill Research Center for Intelligent Machines (CIM) . He directs the McGill Mobile Robotics Laboratory. His sabbatical from 2000-2001 involves stays at Xerox Palo Alto Research Center and Stanford University Dept. of Computer Science.

He obtained his PhD in computer science (computational vision) from the University of Toronto.

He has published some 120 research papers on subjects including visual object description and recognition, robotic navigation and map construction, distributed system design and biological perception. This includes a recent book entitled "Computational Principles of Mobile Robotics" co-authored with Michael Jenkin and published by Cambridge University Press.


bac-coordinators@cs.stanford.edu
Back to the Colloquium Page
Last modified: Wed Oct 11 12:38:06 PST 2000