Kostas Daniilidis, Professor,  Computer and Information Science Department
University of Pennsylvania 

Shape-based object recognition for manipulation and navigation

Thursday, November 17, 4:00 PM

Packard Lab room 466

Reception prior to talk in Packard Lobby

Abstract:  3D objects are characterized by their shape rather than 
their texture or color. Shape is important when objects have to be 
grasped. 3D shapes are now available in databases like the 3D warehouse. 
We represent the 3D object as the set of all its silhouettes as observed
from all viewpoints on a sphere. We study the problem of 3D object
detection and pose estimation as a selection problem in the space of 
segments of the images. We introduce a new shape descriptor the 
chordiogram which is holistic and enables the relaxation of the
 combinatorial selection into a one step convex problem without 
sliding any window for template matching. When the input is video
 the persistence of segments is achieved by a new co-segmentation 
approach in time we called co-saliency. Such a tracking of segments 
constrains the selection problem into the selection of the most 
prominent motion. When objects are used for way finding in 
navigation we show that avoiding hard-commitment on the
detection of objects enables a probabilistic localization approach to
converge even if a map is represented by very few semantically
annotated objects.

Joint work with Alexander Toshev, Alexander Patterson, Roy Anati,
Konstantin Derpanis, Davide Scaramuzza,  Jianbo Shi, and Ben Taskar

Bio: Kostas Daniilidis is Professor of Computer and Information Science at
the University of Pennsylvania where he has been faculty since 1998.
He is the director of the interdisciplinary GRASP laboratory, which
hosts 15 faculty and more than 80 doctoral students and post docs. He
obtained his undergraduate degree in Electrical Engineering from the
National Technical University of Athens, 1986, and his PhD  in
Computer Science from the University of Karlsruhe, 1992. His research
interests are on 3D object recognition, visual motion and navigation,
image matching, stereo vision, and camera design.  He was Associate
Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence 
from 2003 to 2007.  He co-chaired with Pollefeys 3DPVT 2006, and he was 
Program co-chair of ECCV 2010.

© 2012 Computer Science and Engineering, P.C. Rossin College of Engineering & Applied Science, Lehigh University, Bethlehem PA 18015.