• Font Size    
E-mail

Close Window E-mail This Page

CMU Student Develops Picture Location Technology

Required fields are marked with an asterisk(*)



The information you provide will be used only to send the requested e-mail and will not be used to send any other e-mail communications. Read more in our Privacy Policy

Send E-mail

   Print     Share +   

CMU Student Develops Picture Location Technology

Link: CMU.edu

PITTSBURGH (KDKA) ― A Carnegie Mellon University graduate student and his professor are working on some breakthrough technology.

They say they are developing a computer program that would be able to identify the location in a photograph it has never been exposed to before.

The technology is still in its infancy, but the idea is to be able to take any photograph, scan it into a computer and have the computer identify where that photo was taken anywhere in the world.

Experts say this technology could be used for everything from just having fun to solving crimes.

CMU computer science graduate student James Hays says the CBS series "Numb3rs" has written the technology into their show.

"In one of the episodes, in an attempt to track down a killer, they have a photograph of a certain mountain, and they need to know where that location is in the world," says Hays.

In "Numb3rs," it's all fiction, but Hays is developing a computer algorithm to match photographs with known photos and their GPS coordinates.

"A computer is generally not as good as a human at a lot of vision tasks, but computers are good at leveraging lots of data, so what we're doing is harnessing the power of the internet and downloading millions of pictures with known GPS coordinates," says Hays.

At this stage, experts say the system is only as good as the photos already online.

During a test with a shot of the T-line in Dormont, the computer searched the globe and found similar pictures in Italy, England, and Japan.

"It basically worked about 16 percent of the time, one in six times was it able to tell you at the first guess where this image was taken," adds Hays.

A picture from McKeesport and another from Butler County had no local match in the Flickr image database the program uses. But Hays says the program is not yet ready for every photograph in the world.

"More than enough data than we can handle right now," says Hays. "In fact, it's more than two billion images."

So as much as fictional programs like "Numb3rs" would like to think computers can identify the precise location of every photo, computer scientists are working on it but are not quite there yet.

More Information:  IM2GPS

(© MMIX, CBS Broadcasting Inc. All Rights Reserved.)

You need the latest Flash player to view video content.
Click here to download.

Click here to bypass this detection if you already have the latest Flash Player.