Video Acquisition, Processing and Modeling of a Dynamic Scene by Many Cameras
Takeo Kanade
Computer vision technologies, as applied to video media, can do more than just convert images to nicer looking ones. They enable us to extract the information about the scene, such as motion and shape of objects contained in the image sequence. Using a few examples, I will talk about how a more interesting and captivating video of a dynamic scene can be acquired by using a large number of advanced real-time robotics cameras, and about how three-dimensional shape and object motion information can be effectively extracted by using the algebraic structure inherent in the image sequences.
In particular, I will describe the following three examples. Firstly, in 2001 we worked with CBS Sports to develop a multi-robot camera system that was used to broadcast the Super Bowl XXXV. The system, which CBS calls "EyeVision", produces a movie-Matrix-like surrounding view of various interesting and controversial plays. Secondly, the Carnegie Mellon University's Virtualized Reality project has built a 3D room - a fully digital room that can capture an event occurring in it by many (at this moment 50) surrounding video cameras. From those videos, we can obtain a sequence of a complete 3D representation of the event, named 4D representation. A synthetic video of the event from any viewpoint, including those for which no real camera exists can be created. Finally, we can now extract the articulated kinematic data of human limbs, such as limb's length, and track their motion, just like a marker-based motion capture system, but without using any markers. With the extracted kinematics and motion, one's motion can be "transferred" to somebody else's. The talk will consist of many fun video examples.
Short Biography:
Takeo Kanade is U. A. Helen Whitaker University Professor of Computer Science and Robotics at Carnegie Mellon University in Pittsburgh. He received his Doctoral degree in Electrical Engineering from Kyoto University, Japan, in 1974. After holding a faculty position at the Department of Information Science, Kyoto University, he joined Carnegie Mellon University in 1980. At Carnegie Mellon he was Director of the Robotics Institute from 1992 to 2001. Also, he is the Director (part time) of Digital Human Research Center at Advanced Industrial Science and Technology in Tokyo.Dr. Kanade has performed research in multiple areas of robotics: vision, multi-media, manipulators, autonomous mobile robots, and sensors. At Carnegie Mellon, he has led and has been leading successfully many major robotics projects as the principal or co-principal investigator. One of his recent accomplishments was the development of the multi-camera technology that was used for the broadcasting of 2001 Super Bowl IIIV as CBS "EyeVision" - the movie Matrix-like replay of dynamic events.
Dr. Kanade has been elected to the National Academy of Engineering, a Fellow of the IEEE, a Fellow of ACM, and a Founding Fellow of American Association of Artificial Intelligence. He founded and served as the editor of International Journal of Computer Vision, which is regarded as one of the highest-quality journal of the field. Awards he received include the C&C Award, the Joseph Engelberger Award, the Allen Newell Research Excellence Award, the JARA Award, and several best paper awards at international conferences and journals.
Homepage: http://www.ri.cmu.edu/people/kanade_takeo.html