[Cs4800] Fwd: TALK:Thursday 1-21-10 PhD Defense: Body-Relative Navigation Guidance using Uncalibrated Cameras

[Karl Lieberherr]

Another interesting use of graphs. -- Karl

---------- Forwarded message ----------
From: CSAIL Event Calendar <eventcalendar at csail.mit.edu>
Date: Thu, Jan 14, 2010 at 12:39 PM
Subject: TALK:Thursday 1-21-10 PhD Defense: Body-Relative Navigation
Guidance using Uncalibrated Cameras
To: seminars at csail.mit.edu



PhD Defense: Body-Relative Navigation Guidance using Uncalibrated Cameras
Speaker: Olivier Koch
Speaker Affiliation: EECS
Host: Prof. Seth Teller
Host Affiliation: EECS

Date: 1-21-2010
Time: 3:00 PM - 4:00 PM
Refreshments: 4:30 PM
Location: Star (D463)

 The past decade has seen tremendous progress in technologies for
human-oriented navigation, such as GPS-based applications in cars
and on smart phones.  Navigation capabilities are particularly
useful for the blind, for first responders, and for search-and-
rescue teams.  However, GPS reception is either poor or absent
in environments such as buildings, caves, forests and canyons,
thus motivating the need for navigation systems able to operate
using only local sensor data.

This thesis describes a body-worn, vision-based sensor suite and
set of algorithms for human-centered navigation through GPS-denied
environments. We represent the user's motion path as an undirected
graph where nodes represent places and edges represent physical
connections between places. This representation removes the need
for metrical mapping of the environment, while opening the door to
new and challenging problems. Given a graph, the navigation problem
consists of two sub-problems: first, localizing the user within the
graph; second, providing directional guidance to the user given some
target destination.

We demonstrate an end-to-end method for generating the graph, detecting
when the user revisits a place that has been previously observed, and
navigating through the graph. In particular, we present a novel method
that learns to estimate the orientation of the user from feature
matching. We show that our approach can effectively guide untrained
human users through real, unfamiliar environments. We also show an
application of our method to a wheeled robot equipped with a local
obstacle avoidance capability through 2km of indoor excursion.

Relevant URL(S): http://people.csail.mit.edu/koch
For more information please contact: Olivier Koch, 6174477421,
koch at csail.mit.edu

_______________________________________________
Seminars mailing list
Seminars at lists.csail.mit.edu
https://lists.csail.mit.edu/mailman/listinfo/seminars
-------------- next part --------------
HTML attachment scrubbed and removed