Autodesk partnered with Cambridge, Mass.-based Smartvid.io to develop technology that will scan construction-site photos and video to spot issues that can affect performance, quality and safety.
Using artificial intelligence to scan construction-site photos and video to spot issues that can affect performance, quality and safety, a tech startup has landed a strategic investor with millions of such images that can be used to improve the technology. Under the agreement, the startup reserves its right to continue integrating its tool with the products of other developers.
Cambridge, Mass.-based Smartvid.io is getting a $7-million cash infusion from its existing investors as well as Autodesk Inc.; Borealis Ventures, a venture capital fund focused on technology for construction; and another venture capital firm, Castor Ventures. The deal also gives the company, whose service already is integrated with Autodesk BIM360 Field, the ability to use Autodesk’s trove of construction images for improving its image recognition technology.
“Strategic investors can bring not only money. They also can bring data to the play,” says Josh Kanner, founder and CEO of Smartvid.io. “There are millions of photos in the Autodesk BIM 360 platform that can be analyzed using deep learning,” a form of artificial intelligence.
Kanner should know. One of his prior startups, Vela Systems, was purchased by Autodesk in 2012 and rebranded BIM360 Field.
Smartvid.io’s analysis engine, called “Vinnie,” uses deep learning. The system takes its cues from the way human brains link networks of neurons to gather details and draw conclusions by refining data.
Incremental conclusions are given a probability of truth before being passed along for refinement by the next level of analysis. With the rapidly improving power of today’s processors, dozens and even hundreds of such layers of analysis can be nested to refine conclusions about image content.
“The first step in looking for safety hazards is to find people. Then, you look within that area for safety colors or hardhats,” says Kanner. The same approach can be used to find concrete cracks to judge their significance. The more times Vinnie correctly analyzes image details, the more accurate it becomes, which makes having an enormous cache of images for training an especially valuable resource.
In its early development, Vinnie was given access to several years of ENR Photo Contest submissions for initial training. “It really starts with the data,” says Kanner.
Sarah Hodges, director of Autodesk’s construction business line, says Autodesk and Kanner have a shared interest in and appreciation for nested analysis, which is at the heart of Autodesk’s own AI research-and-development initiative, called Project IQ. “It’s a very strategic partnership,” she says. “It’s not just for the now. It’s really about the more long-term strategic gains we can provide to our customers.”