We all know Google Earth, the tech giant’s app that lets users look at pretty much any spot on the planet via high-res satellite photography. It’s a brilliant piece of technology—the fact that it even exists is fairly astounding, the fact that it works so well is positively mind-boggling. Now, researchers at Carnegie Mellon University in Pittsburgh, Pa., have developed a new add-on that can spot patterns in the landscape and collect similar images. It’s on-the-nose name: Terrapattern.
Get It? Patterns In the Terrain
Satellite images of nearly any visually-recognizable pattern can be quickly spotted and compiled using Terrapattern. Want to see images of all the tennis courts in the area? You got! Pics of traffic roundabouts? No problem! Even boat wakes caught by Google Earth’s satellite cameras can be found.
So just how the heck does this incredible new technology work? It’s a sort of modified A.I. applied to satellite imagery, Carnegie Mellon’s researchers say. The team started with a deep convolutional neural network (say that five times fast!), the kind of A.I. that’s often used for image-recognition. They then refined it through ImageNet, an online database of over a million, presorted images, which helped Terrapattern begin to identify patterns in the landscape.
From there, Terrapattern was connected to OpenStreetMap—a kind of Wikipedia world map, complete with volunteer-provided identification of streets, rivers, parks, and the like. Four days later, Terrapattern’s A.I. had developed its own way of identifying the colors, shapes, contrasts, and textures that make up the patterns it uses to spot like terrain.
Small Budget, Big Success
Surprisingly—maybe even shockingly—Terrapattern was developed from start to finish on a grant of just $35,000. Given that most complex technology of this kind can cost orders of magnitude more, it’s amazing that Terrapattern even exists in a finished form, let alone that it works so well.
“It actually does the thing we wanted it to do,” said Golan Levin, associate professor of art at Carnegie Mellon and one of the lead developers of the technology. “We built Terrapattern to allow people to search the world according to how they see it. But it also puts them in dialogue with the landscape as seen by machines. And that’s just a very uncanny experience.”
At this time, Terrapattern is limited to Pittsburgh, San Francisco, and New York. You can try it for yourself right here.