The Atlantic digs into Google Maps' Ground Truth

Google Maps has gotten a lot of praise lately, in part because the Apple Maps app that replaced it in iOS 6 has been somewhat of a disaster. We've discussed the Apple app's shortcomings already, but what about what makes the Google alternative so good? The Atlantic has posted a behind-the-scenes look at how Google builds its maps, providing some insight on the answer.

One of the keys is the Ground Truth, a base map that has much more information than is presented to users. This foundation combines data from multiple sources, including satellite imagery, geological surveys, and Street View.

The contributions of Google's camera-equipped cars seem to be particularly valuable. First, there's the volume. Right now, Street View reportedly adds more imagery data in two weeks than Google itself possessed in 2006. Google runs algorithms on those images to look for traffic signs, addresses, and identifiable corporate logos like McDonalds' golden arches. All that information is fed into the Ground Truth to create more accurate maps. The Atlantic likens Street View to web-crawling in the real world, which makes perfect sense given Google's roots.

According to the article, compiling all the mapping data for a single country takes hundreds of people. There seems to be lots of hand-tuning involved. Google Maps isn't immune to inaccuracies, though. "Several thousand" errors are reported daily by end users, and the team dedicated to addressing them strives to correct problems within minutes. That team seems to be doing its job, since I've only encountered a handful of Google Maps errors in years of using the service.

For me, the integration of data derived from Street View imagery is by far the most intriguing element of Google's mapping process. Although the street-level view may be used rarely, the data it contributes could be invaluable to maintaining Google's lead in the mapping field.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.