Computer Vision Distinguished Seminar Series
Detecting and Recognizing Objects in the Real World
Dr. Jan-Olof Eklundh
Thursday, January 10, 2008
2:00PM ~ 3:30PM, Harris Center 101
Abstract
In our work we are interested in performing similar tasks on systems looking at a real scene, such as a robot looking for objects in an indoor environment. The question this raises is if existing methods apply directly, is it possible to take developed appearance based and statistical techniques and use them as they are? Our findings indicate that this is not always the case, at least not if one wants the system to behave robustly in a realistic environment.
Our talk will focus on a number of pertinent problems from the point of view of an integrated system: figure-ground segmentation and selecting appropriate views, bottom-up and top-down attention, the role of 3D cues, and the use of multiple cues. In addition we will discuss the close ties bewteen perception and action which become evident when studying vision from such a systems point of view and in particular when talking about object classification.
Short Bio
Eklundh graduated in mathematics from Stockholm University in 1970. He then joined the newly formed Laboratory for Image Analysis at the National Defense Research Institute, Stockholm and spent 1977-79 at the Computer Vision Lab, University of Maryland. In 1982, he became associate professor at KTH where he founded the Computer Vision & Active Perception Lab, CVAP. In 1996, he initiated the Center for Autonomous Systems, in which CVAP is now a key partner. In 1986, he became professor in computer science and was Dean of the School of Electrical Engineering and Information Technology at KTH from 1995- 2001.
His research interests cover a broad range of topics in computational vision, image processing, and robotics, especially active visual machine perception with relations to human vision, analysis of shape and geometry and motion, behavioral aspects of perception, and perceptually guided autonomous systems.
