IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

New AI Tool Lets Government Identify Objects from the Air

The Australian company Nearmap, which also operates in the U.S., has accumulated a catalog of high-resolution aerial images for 90 million parcels, and now it’s applying AI to learn when property owners make changes.

Georgia capitol Nearmap aerial photo
An aerial photo of the Georgia state capitol building taken by Nearmap.
In the span of a few decades, widespread geospatial imagery has gone from impractical to unremarkable, and governments are still finding uses for it. From covert surveillance to disaster assessment, high-resolution aerial photos have become one of many regular data assets used by larger agencies to study problems or changes to property.

Correspondingly, aerial imagery has also become a more competitive market, and the Australian company Nearmap this week made a move in hopes of distinguishing itself by adding artificial intelligence to its offering to make property changes easier to flag and identify.

As described in the company’s news release, Nearmap AI is now available in the company’s Mapbrowser tool, in the form of different layers atop regularly updated images catalogued by the platform. These layers can automatically identify and analyze ground features such as residential building footprints, construction sites, tree overhang, green cover, solar panels, trampolines and swimming pools, and detect changes over time.

Founded in 2008 and expanded to the U.S. in 2014, Nearmap has been building a massive database of aerial imagery for years, flying planes at 12,000 feet over both Australia and North America that  take photographs at 3-inch resolution — in which one pixel is about 3 inches wide. Nearmap's CEO, Rob Newman, said they’ve covered land that contains about 72 percent of the U.S. population, including the 415 largest cities in North America, and made the imagery accessible on the cloud by subscription.

Newman said government agencies and insurance companies have been using the imagery to learn about changes to infrastructure or property, but several years ago they started asking if there was an easier way than visually scanning through it themselves. Newman said he put together a research team of 20 data scientists in 2017 to study the problem, and they produced an AI tool.

Newman pointed out the alternative is using satellite imagery, which doesn’t have sufficient resolution; or sending an employee in a car to take photos, in which case he would take more time and likely wouldn’t be able to see the back yard.

“A local government could say, ‘Give me a listing of every swimming pool in my county,’ they click the button, and it exports out immediately to them a complete listing by address, where all the swimming pools are in that county,” said Newman. “Some of the features that you might [use] to evaluate property are difficult to do from low-resolution imagery. Using Nearmap imagery, we can very quickly see things. ... Using AI almost automates that for you, showing you every house that’s got a swimming pool, and that’ll be highlighted in a visual way. You’ll get a bright blue dollar [sign] for every swimming pool, and you can then export that data out and put it into your systems.”

As cameras and cloud technology have become more competitive, so has the business of selling subscriptions to aerial imagery. Earlier this year, one of the market’s domestic players, Vexcel Imaging in Colorado, acquired a fleet of planes and other assets from Geomni, a subsidiary of risk analytics company Verisk, with the goal of building the world’s largest library of geospatial imagery.

Nearmap would be a competitor in that endeavor, regularly updating its catalogue of 90 million parcels and more than 380,000 square miles across the U.S. and Australia, according to its news release. Newman said that reach, and AI, is what distinguishes his company from Vexcel.

“Having 90 million parcels globally, processed through our AI engine, is something that nobody’s been able to achieve at scale like we have,” he said.

Editor's note: This story has been updated to clarify that the number of square miles Nearmap covers is in multiple countries.
Andrew Westrope is managing editor of the Center for Digital Education. Before that, he was a staff writer for Government Technology, and previously was a reporter and editor at community newspapers. He has a bachelor’s degree in physiology from Michigan State University and lives in Northern California.