As more than a billion people have come to rely on Google Maps to explore the world and millions of apps and experiences have been built on top of our data, we’re often asked how we build the map that serves such a wide set of users and use cases.
The answer is that it’s taken more than a decade of laying the groundwork and an obsessive commitment to refining our techniques to be able to meet increasing user expectations for fresh and accurate data and insights.
An early investment in imagery
Just a couple of years after launching Google Maps and Google Maps Platform (formerly Google Maps APIs), we launched Street View. For consumers, it helped them virtually explore the entire world from their own homes. For as long as our Street View program has operated, we’ve made this rich imagery data set available to businesses so they can provide real-world context in their applications. Our Street View APIs allow real estate sites like Trulia to help homebuyers discover a place they’ll love to live by virtually exploring neighborhoods right from their website and apps.
At Google, Street View gave us the foundation for the future of our mapping process. Advances in our machine learning technology, combined with the more than 170 billion Street View images across 87 countries, enable us to automate the extraction of information from those images and keep data like street names, addresses, and business names up to date for our customers. If a picture is worth a thousand words, then a high-res, panoramic image is worth a billion. So we’re committed to developing our own hardware, like our newest trekker equipped with higher-resolution sensors and an increased aperture, to deliver the highest quality imagery and insights to our customers.
Partnering with authoritative sources
Providing reliable and up-to-date information is essential for enterprises looking to build mission critical applications on our platform. So we also use data from more than 1,000 authoritative data sources around the world like the United States Geological Survey, the National Institute of Statistics and Geography (INEGI) in Mexico, local municipalities, and even housing developers.
Combining our imagery analysis with third-party data gives customers the most accurate and reliable data to power their businesses. For instance, we’re able to provide ridesharing companies such as Lyft, and mytaxi with convenient pickup/dropoff locations for their passengers and traffic-aware routing so their drivers can take the fastest route possible. We understand that one wrong route or delayed pick-up can have an impact on whether a customer comes back, so we make it easy for third-party authoritative sources to share their data with us. From there, we quickly ingest it and turn it into the features that are helping ridesharing companies all over the world improve their customer experience and business efficiencies..
Real people, real insights
Data and imagery are key components of mapmaking. But they’re static and don’t always give us the context we want about a specific place. If you think of Street View as helping you contextualize where you are on a street, you can think about user contributed content as helping you contextualize a specific place like a restaurant or coffee shop. With the help of a passionate community of Local Guides, active Google users, and business owners via Google My Business, we receive more than 20 million contributions from users every day–from road closures, to details about a place’s atmosphere, to new businesses, and more. To ensure this contributed info is helpful, we publish it only if we have a high degree of confidence in its accuracy.
This has enabled us to build a data set of more than 150 million places around the world, which we make available to developers through our Places API. The Places API includes rich data on location names, addresses, ratings, reviews, contact information, business hours, and atmosphere–helping companies empower their users not just to find a restaurant, but to find a restaurant that’s good for kids with vegetarian menu items.
Keeping up with the speed of innovation and growth with machine learning
The mapmaking process we’ve shared so far builds a useful and reliable map, but it presents one major challenge–speed. To empower our customers to move fast and innovate, we need to map the world more quickly than ever before. And as regions of the world rapidly develop, we need to be able to quickly get that information into our maps and products. To increase the rate at which we map the world, we turn to machine learning to automate mapping processes, while maintaining high levels of accuracy and precision.
Here’s an example of how we used machine learning to solve what we dubbed “fuzzy buildings”. Our team was frustrated with fuzzy building outlines caused by an algorithm that tried to guess whether part of an image was a building or not. To fix this, we worked with our data operations team to trace common building outlines manually. Now that’s a solution in itself. But tracing all the common building outlines in the world by hand isn’t a scalable or quick process. So once our team traced the common building outlines, they used this information to teach our machine learning algorithms what shapes buildings tend to have in the real world, and which parts of images correspond with building edges and outlines. Using this technique, we were able to map as many buildings in one year as we mapped in the previous 10–vastly improving the maps we share with our customers.