By Russell Steinthal

Wouldn't it be great to have eyes in the back of your head, to be able to see everything around you without turning your head? Biology may have made that impossible, but the newly developed OMNICAM system currently on display in the fourth floor lobby of Mudd may be the next best thing.

The omnidirectional video camera, developed by a team from the Computer Vision Library led by Associate Professor of Computer Science Shree K. Nayar, generates an image of the entire 360 sphere around the camera without moving the camera at all.

That omnidirectional image is then fed to OMNIVIDEO, a computer program written by Venkat Peri, a graduate student in electrical engineering, which allows the user to display distortion-free images of any point on in the camera's sphere of vision at variable magnifications.

According to Professor Nayar, the theoretical problem at the heart of the OMNICAM sensor is "to grab rays of light from different parts of the scene and reflect them through the mirror so that they intersect at a single point."

Existing omnidirectional cameras that operate though a fisheye lens or through the use of multiple cameras cannot achieve that goal since they produce multiple centers of projection. In contrast, OMNICAM uses a specially designed and built parabolic mirror, which along with the OMNIVIDEO software, are the only custom components in the entire system.

Professor Nayar highlighted three main uses for the innovative camera: teleconferencing, video surveillance, and remote exploration. The first two take advantage of OMNIVIDEO's ability to generate multiple real-time images from the same video signal from the OMNICAM.

For example, one OMNICAM positioned in the middle of a conference table could be setup to provide individual images of each person seated around the table.

As an example of remote exploration, Professor Nayar envisions potentially placing an OMNICAM on the roof of a building and allowing users to scan the New York City skyline, possibly providing a short textual description of the area being viewed. While there are still issues to be resolved before that plan can be implemented, he said that he expects it to be implemented in the future.

Email the Author

Columbia Computer Science Department

Professor Shree K. Nayar

© Copyright 1996 Columbia University