Search This Blog

Monday, June 29, 2015

Paired with AI and VR, Google Earth Could Change the Planet

As reported by Wired: THE JAMES RESERVE is a place where the natural meets the digital.

Part of the San Jacinto mountain range in Southern California, the James is a nature reserve that covers nearly 30 acres. It’s closed to the public. It’s off the grid. Vehicles aren’t allowed. But Sean Askay calls it “one of the most heavily instrumented places in the US.” Robots on high-tension cables drop climate sensors into this high-altitude forest. Bird’s nests include automated cameras and their own sensors. Overseen by the University of California, Riverside, the reserve doubles as a research field station for biologists, academics, and commercial scientists.

In 2005, as a master’s student at the university, Askay took the experiment further still, using Google Earth to create a visual interface for all those cameras and sensors. “Basically, I built a virtual representation of the entire reserve,” he says. “You could ‘fly in’ and look at live video feeds or temperature graphs from inside a bird box.”

Somewhere along the way, the project caught the eye of Google’s Vint Cerf, a founding father of the Internet, and in 2007, Askay moved to Mountain View, California, home to Google headquarters. There, he joined the team that ran Google Earth, a sweeping software service that blends satellite photos and other images to create a digital window onto our planet (and other celestial bodies). Since joining the company, the 36-year-old has used the tool to build maps of war casualties in Iraq and Afghanistan. He put the service on the International Space Station, so astronauts could better understand where they were. Working alongside Buzz Aldrin, he built a digital tour of the Apollo 11 moon landing.

Now, as Google Earth celebrates its 10th anniversary, Askay is taking over the entire project—as lead engineer—following the departure of founder Brian McClendon. He takes over at a time when the service is poised to evolve into a far more powerful research tool, an enormous echo of his work at the James Reserve. When it debuted in 2005, Google Earth was a wonderfully intriguing novelty. From your personal computer, you could zoom in on the roof of your house or get a bird’s eye view of the park where you made out with your first girlfriend. But it proved to be more than just a party trick. And with the rapid rise of two other digital technologies—neural networks and virtual reality—the possibilities will only expand.
Click to Open Overlay Gallery
Through an extension to Google’s Chrome web browser called Earth View, you can view “the most beautiful and striking” satellite images from around the world, “diving in” to places like Cuba. GOOGLE

A Visit to Prague
Neural networks—vast networks of machines that mimic the web of neurons in the human brain—can scour Google Earth in search of deforestation. They can track agricultural crops across the globe in an effort to identify future food shortages. They can examine the world’s oil tankers in an effort to predict gas prices. And it so happens that Google runs one of the most advanced neural networking operations in the world. For Google Earth, Askay says, “machine learning is the next frontier.”

According to Askay’s boss, Rebecca Moore, the company is already using neural networks to examine Google’s vast trove of satellite imagery. “We have the Google Brain,” she says, referring to the central neural networking operation Google has built inside the company, “and we’re doing some experiments.” That’s news. But it’s not that surprising. Two startups—Orbital Insight and Descartes Labs—are already doing much the same thing.

Meanwhile, virtual reality—as exhibited by headsets like Facebook’s Oculus Rift and Google Cardboard—is bringing a new level of fidelity and, indeed, realism to the kind of immersive digital experience offered by Google Earth. Today, using satellite imagery and street-level photos, Askay and Google are already building 3-D models of real-life places like Prague that you can visit from your desktop PC (see video at top). But in the near future, this experience will move into Oculus-like headsets, which can make you feel like you’re really there.

“We have so much interesting stuff,” Askay says of Google Earth’s massive collection of images. “How amazing would it be to experience Google Earth in that environment?”

The Google Outreach
Google isn’t the only one that will drive the evolution of Google Earth. In 2007, not long after taking the job at Google, Askay flew to Brazil, helping an indigenous tribe, the Surui, map deforestation in their area of the Amazon, and this gave rise to a wider project called Google Earth Engine. With Earth Engine, outside developers and companies can use Google’s enormous network of data centers to run sweeping calculations on the company’s satellite imagery and other environmental data, a digital catalog that dates back more than 40 years.

“So, if you want to look at 40 years of Landsat imagery and do change projection over time, you can,” Askay says. “You could do retrospective models of where deforestation took place and how fast, as well as predictive models and even near real-time detection. We’re getting to the point where we can start sending alerts saying that something that looks like deforestation has occurred in the last three days.”

As it stands, Earth Engine is only available to a limited number of outsiders, but Askay and Moore say Google plans to gradually open it up to a much larger audience. With a project called Map of Life, independent researchers are already examining how global warming in changing the habitat ranges of particular animal species. Others are working to track water resources. The World Resources Institute now uses the service to provide a map of deforestation not only in the Amazon, but across the globe.

Learning Gets Deep
At the moment, this map is generated with traditional computing techniques. But according to WIR chief technology officer Aaron Steele, the organization is now building a system that can expand, accelerate, and improve the process with neural networking. At companies like Google and Facebook, “deep learning” has already recognizing faces and objects in online photos uploaded by everyday internet users, and many believe the technology can significantly accelerate the analysis of satellite imagery—and to much greater effect.

“It’s a real breakthrough in computer vision, though in order to make it work, you need a whole lot of compute power,” says Steven Brumby, a former Los Alamos National Lab researcher and co-founder of Descartes Labs. “The opportunity is enormous.”

In a recent months, Google has shown how effective this AI technology can be in analyzing ground-level photos captured as part of its Street View project. In building its online maps, Google once used human editors to identify addresses in photos. Now, neural networking does this automatically.

James Crawford, an ex-Googler and the CEO of Orbital Insights, a company dedicated to examining satellite imagery with neural networks, agrees that Google Earth is ripe for this kind of automatic analysis. But after using Google Earth Engine as an outside developer, he says the service may need to evolve further before it’s suitable to such work. “The level of control we need in our pipeline, it didn’t really work for us,” he says. “But that could change.”

The Cardboard Effect
Elsewhere at Google, an eclectic team of engineers are building a virtual reality system for use with headsets that strap around the eyes. It began with a “20 percent” project called Google Cardboard, a headset made out of cardboard—literally—that works in tandem with your smartphone. In the beginning, this seemed like an ironic comment on the wider push toward virtual reality. But it has evolved into something far more serious.

These engineers have designed 16-lens cameras that can capture 360-degree stereoscopic images of the real world. GoPro will soon offer these cameras to the world at large. And Google is offering a service that can stitch those images into a 360-degree digital environment for viewing in Cardboard and, potentially, other headsets. “We have ambitions beyond just Cardboard,” project leader Clay Bavor told us last month. “There are many other things going on.”

The project is natural fit for Google Earth. The Cardboard team recently unveiled what it calls Google Expeditions, a way for school students to experience distant lands via the headset, and the concept is that much more attractive in the context of the earth as a whole. “You can imagine this as a Google Earth experience,” Askay says.

For Moore, this kind of thing isn’t a novelty. It’s education. “It’s not just for fun and gaming,” she says. “It can give people a more immersive understanding of the planet—places that matter and places that are changing.”
Street View now includes “special collections” that show off places like the Great Barrier Reef. GOOGLE

A New GlobeIt’s also worth remembering that Google now has its own satellite company. Last summer, it acquired Skybox, a startup that uses cube satellites to take more frequent and higher resolution photos from the skies. According to Askay and Moore, Google hasn’t yet incorporated the Skybox imagery into Google Earth. But this will come.

On one level, the prospect of pairing Skybox with technology such as neural networks is a frightening thing—another erosion of privacy in the physical world. But it also brings enormous possibilities. Today, Google Earth is a nice way to look at the planet—not to the mention Mars, the Moon, and the heavens. But in the years to come, it will grow into something else. Virtual reality will bring new fidelity. And AI and other types of data analysis will bring a new understanding our planet.

“What I’m looking forward to is combining Google Earth with the kind of dynamic data coming out of Earth Engine—data on deforestation, floods, temperatures,” Moore says. “If you render that kind of information on Google Earth, it becomes a living, breathing dashboard of the planet. You can put in everyone’s hands, not just charts and graphs of what’s going on, but high-resolution information that’s sitting, almost literally, on the surface of the earth.” It’s like Askay’s work at the James Reserve. But on much larger scale.