DO YOU fancy blasting your boss to oblivion as you play your favourite
computer game? Then engineers in the Visual Geometry Group at Oxford University
have just the thing: software that converts ordinary, flat images into a
high-resolution 3D graphic that you can insert into a video game.
Using a single camera, such as a computer-mounted webcam, the software
captures a succession of images from an object moving across its field of view.
From this it can map the object鈥檚 precise shape. It requires no information
other than the images themselves, says Andrew Fitzgibbon, one of the developers
at Oxford.
Some previous systems that produced a similar result had to use several
cameras at different angles. Others used range-finding lasers, or required
special markers or patterns to be placed on the object so that its movement
could be tracked accurately.
Advertisement
The Oxford software compares each video frame with the following two, looking
for points of reference such as lines, corners and curves. It can use 鈥渁nything
that鈥檚 different from its surroundings鈥, says Geoff Cross, one of the
researchers. 鈥淵ou don鈥檛 have to `know鈥 the motion at all. You can get that from
the way the object moves in the image.鈥 To do this effectively the software
exploits techniques that people use to glean motion from images. For example, if
a point on an object is moving to the left, then it is usually safe to say that
most of the other points in its vicinity will be as well. Once the motion is
known, the shape of the object, or at least part of its surface such as
someone鈥檚 face, can be mapped by triangulation of the three images as if they
came from different cameras. Cross admits that this could make it tricky
capturing someone鈥檚 face covertly鈥攚hich would in any case throw up many
privacy issues鈥攂ut it could work by having someone sit at the computer
terminal with a webcam aimed at them, ensuring that at least the face has been
captured.
Cross says that games such as Quake already have a feature that lets users
customise characters, and tools are available on the Internet that let you draw
them. But you can鈥檛 beat an image of a real person. And because the Oxford
software is relatively simple and requires very little memory, Cross hopes that
games publishers will find it easy to include it in their programs.