All Google needs to update business info is a Street View photo
Google is no stranger to using machine learning to improve its products — or save manatees. To that end, the internet juggernaut has announced that its algorithms are capable of successfully pulling business names and phone numbers from Street View photos. In its tests, the technology was successful at “reading” French street signs over 84 percent of the time. Meaning, now a Street View car can roam the streets of a city and fill in a business’ Google Maps profile automatically. It stems from Google’s work using machine learning and computer vision to blur out faces and license plates.
More directly, it builds on the way that Google has been extracting house numbers for Street View for the past few years. It took a lot of doing, though. Datasets made up of multiple versions of the same signs had to be pored over to make up for visual artifacts and blurring. The algorithms then pieced these images together to extract the names and data from them.
Google says that this has been used to improve the location data for around a third of addresses around the world. Perhaps even on the Skyfall island and in Middle Zealand. If you fancy using Google’s dataset for yourself, the framework is available on GitHub.
Source: GitHub, arXiv, Google Research