Google Street View Employs AI To Read Street, Business Signs

Google Street View is all grown up. For the first time in nearly eight years, its cameras are finally getting an upgrade. But more than just producing clearer and crisper images, the higher quality images from the more powerful hardware have another purpose. In line with Google's favorite obsession, Google Street View is putting machine learning and artificial intelligence to work, using these high res images to detect, process, and, if necessary, translate street names and business signs, making it even easier to index the whole, physical world and put them in Google's hands.

Street View has been to all sorts of places where very few of us can even dream of reaching. And, no, we're not just referring to exotic vacation spots. Under the sea, near a volcano, and even in outer space, Street View has been there so that we won't have to. And yet, until recently, it has been an excruciating and partly manual process.

Google is the master of making computers do the work instead of inefficient and error-prone humans. While that is fine for things like text or even voice search, indexing the physical world is a slightly different and more complicated affair. Fortunately, Google has also been building its library of computer vision algorithms and techniques, which t can put to good use in Street View.

But in order to analyze those Street View images, Google needs better images in the first place. That's why it has employed the expertise of Steve Silverman, who has helped build cameras for two of NASA's Mars rovers. This resulted in a significant reduction in the number of cameras on a custom-purpose Hyundai Street View car, from 15 down to 7 cameras, each with a 20 megapixel sensor.

But just as important are the two side-facing HD cameras whose sole purpose is to capture images of human readable signs and names. These are then fed to Google's machine learning computers. More than just interpreting image to text, Google's algorithms also weed out unnecessary information, like ads or slogans, and interpolate abbreviations into their full street names, for example. This, together with user-submitted 360-degree videos, help keep Google's mapping platform at the top, even while rivals like Microsoft and Apple try to nibble at a tiny part of the Google Maps pie.

SOURCE: WIRED