Augmented-reality machine works in real time
FEBRUARY 8--Computer-generated scenery can be realistically added to live video footage using a machine-vision system developed at Oxford University, UK.
FEBRUARY 8--Computer-generated scenery can be realistically added to live video footage using a machine-vision system developed at Oxford University, UK. Researchers Andrew Davison and Ian Reid say the augmented-reality system could also, in the future, enable robots to navigate more effectively or could be used to virtually decorate a real house or plan engineering work. It allows a computer to build an accurate three-dimensional model of the world using only a video-camera feed. It can also keep track of the camera's movement within its environment--all in real time.
Previously, it has been necessary to calibrate a computer using several markers added to a scene. The Oxford team's machine only requires an object of known size to be placed in its line of sight to perform a complete calibration. The system then automatically picks out its own visual markers from a scene. By measuring the way these markers move, the computer can judge how far away each marker is. It can also rapidly determine how the camera is moving.
Davison says the key is to efficiently search for visual markers. "The system is very selective about when it looks for landmarks in its environment," he said.
John Illingworth, an expert in computer vision at the University of Surrey, UK, says the system could represent an important step toward machine vision. He adds that robots could use the technique to better understand and navigate through a given environment. This might even enable them to act in a more human manner. "It's exciting work," Illingworth said. "One of the reasons we can do so many things is that we know where we are in relation to other things. And to give that ability to machines is quite a challenge--people have been trying to do it for awhile."
In January the project was awarded £255,000 ($480,000) of funding from the UK government Engineering and Physical Sciences Research Council.
More information is available at www.newscientist.com/article.ns?id=dn6965.