Thursday, March 08, 2007
10:00 AM
CSB 601
Deva Ramanan
Toyota Technological Institute, Chicago
Training a Computer to See People
One of the great, open challenges in machine vision is to build a system that can "see people." A reliable solution opens up tremendous possibilities that impact everyday life---from how we drive cars and browse photos to how society cares for medical patients. Such systems pose serious technical challenges from an algorithmic point-of-view. This is because of complex appearance phenomena due to clothing, the articulation of limbs, and the overall complexity of real-world imagery. In this presentation, I will discuss a family of algorithms for understanding people through real-world video and image data.

I will argue that some general techniques emerge from this pool of algorithms. One is the notion of video-for-learning. I describe a tracker that tracks people while simultaneously learning clothing models for individuals in a video. This link between tracking and model-building suggests new learning algorithms for object recognition. Another notion applicable to both video and images is that of integrated visual parsing. By taking a unified approach to parsing, low-level tasks such as edge detection are performed jointly with high-level tasks such as posture estimation. Such an integrated scheme is computationally demanding and so requires some specialized algorithmic tools.

Throughout the talk, I will include various demonstrations on real world imagery including online photo collections, unscripted public footage, a feature-length film, and historic sports footage (from the 2002 World Series and 1998 Winter Olympics).

Bio: Deva Ramanan received the BCE degree with distinction, summa cum laude, from the University of Delaware in 2000. He received the PhD degree in electrical engineering and computer science (EECS) with a designated emphasis in communications, computation, and statistics from the University of California at Berkeley in 2005. He is currently a research assistant professor at the Toyota Technological Institute in Chicago. In the summer of 2004, he was an academic visitor with the Visual Geometry Group at Oxford University. In the summer of 2006, he was a visiting researcher with the Robotics Institute at Carnegie Mellon University. His research has been supported by a University of California MICRO Fellowship and by a US National Science Foundation Graduate Research Fellowship. His interests span computer vision,machine learning, and computer graphics. His graduate work focused on tracking people and recognizing their activities.