understanding what the user is searching for is only half the battle for Google Maps. Once the service understands where the user wants to go, the next challenge is to find the easiest routes to get them there, whether by car, foot, or public transit.
So how does Google Maps figure out the best route for a user to take? How does it know which streets are marked ‘one way’ only, where there are left turn restrictions and which roads are separated by dividers or streetcar tracks?
The attempt to answer these questions led Brian McClendon, vice-president of Google Earth and Google Maps, and his team at the Mountain View, Calif.-based search titan, to launch Ground Truth in 2008, built on the back of the company’s Street View project.
“In 2008, we were in a situation where we were licensing third-party data and organizing it with search as best we could, but we weren’t able to do what we wanted with the data,” Mr. McClendon said in an interview.
“So we made a fundamental decision to go out and make our own maps and have our own data and be able to improve it in terms of both quality and detail. This has been a huge investment for us, focusing on both the data and the search quality.”
To create its digital maps, Google relies on all sorts of data from third-party providers, from satellite imagery and census data to bike trails from riding associations. With the Ground Truth project, Google has steadily been incorporating GPS data and images collected by its fleet of Street View cars, which have covered more than five million unique miles of roads in 3,000 cities in 40 countries.
With Ground Truth, Google has been busy teaching its computers to comb through its vast collection of Street View photos and has developed technology which enables them to recognize street signs, turn restrictions and the locations of businesses. Google combines that information with GPS data to help align the road information.
“It’s actually good GPS data and we make it much better by using computer vision to connect different photos to each other,” Mr. McClendon said. “We then run these computer vision algorithms to do extraction of street signs, speed limit signs, addresses and business logos, and then run these through automated recognition.”
These data can be uploaded immediately to the company’s road database, but if there’s a discrepancy, Google will send a Street View car operator to confirm what’s in the photos.
“The combination of heavy automation, with final human verification of the ambiguous pieces gives us a very high-quality data set that is derived from Street View imagery,” Mr. McClendon said.
No comments:
Post a Comment