A new class of company is growing, one which uses large-scale imagery to predict retail or environmental trends, but doesn’t own the drones and satellites which take the images. Jimi Crawford is the CEO of Orbital Insight, a market intelligence company leveraging information from satellite imagery. He worked at Google Books and NASA’s Ames Research Center before starting Orbital Insight, which raised $8.7 million in Series A funding in March of this year. Market research company CB Insights spoke to Crawford about his company and his views on current technology startups on the first episode of their new podcast, hosted by Michael Dempsey.
Dempsey said that he is seeing more and more market intelligence startups choose to use data from drones and satellites, but not to own the hardware. Orbital Insight uses that big data and satellite imagery to predict global retail trends by counting cars in parking lots, or to track world-wide effects like deforestation.
Companies like this have proliferated in part because of the increased availability and decreased cost of satellites, Crawford said. After the images have been acquired, they are processed on GPUs and run through imagine recognition software enabled by deep learning and convolutional neural networks.
GPUs allow Orbital Insight to process images 500 to 1,000 times faster than they could with CPUs, Crawford said. Part of that processing power comes from a source outside of the industry, Crawford quipped: “So basically, thanks to all the guys playing first person shooter games for funding the development of GPUs, and now we use them to go the other way and process imagery instead of create imagery.”
“There’s been a complete sea change over the last couple of years from where computers were much worse than people at doing image recognition,” Crawford said, “and now computers are on par in many fields, and all due to advances in, basically, neural network design and deep learning.”
Orbital Insight doesn’t plan to change this business model of non-ownership, Crawford said. For one, it fosters a positive relationship with the satellite companies with which they partner. Orbital Insight uses an image for its own purposes – counting cars, for example – and then deletes it.
“So we’re not actually taking away the value they could get from selling that image to somebody else,” said Crawford.
On the other hand, Orbital Insight is in a position to do image analysis at a scale that would cost hedge funds, their major customer base, millions of dollars to do in-house.
Crawford hopes to see the satellite imagery industry leverage more drone footage too. Autonomous, solar-powered drones could orbit single cities at 60,000 feet, tracking shipping or retail. A couple companies have already demonstrated this, he said, but commercial leverage is still a year or two away. The drones would not take images at a high enough resolution to count people, he said, so it could avoid any privacy issues.
They could be used to fill in the gaps between satellites, which typically orbit over their target area during midday. Since they are cheaper than satellites, drones could also be used in the industry to work on projects restricted to specific times and specific places, gathering more specific data for this burgeoning industry.
More from Cawford, including details about deep learning and his opinions on Hyperloop, can be found at CB Insights.
Filed Under: Aerospace + defense