When most people hear terms like “computer vision” and “smart cameras,” they probably think of a few widely discussed uses in emerging fields like facial recognition and self-driving cars.
But the technology also can help existing industries, and that’s the area where the Vancouver-based startup DeepCanopy hopes to make its mark.
“If you can see it, there’s a very good chance you can teach a piece of software to ID it,” says DeepCanopy President and CEO Nate Fuller. “(Computer vision and smart cameras) are still very nascent technologies, so there’s a large opportunity for understanding how they can benefit the industry.”
Many companies are working to develop new applications for computer vision in areas such as manufacturing and production, according to DeepCanopy co-founder and Chief Operating Officer Jonathan Parnell, but the idea for DeepCanopy arose when the founders realized that one critical area appeared to be overlooked.
“No one was really going after the safety monitoring space,” he says. “(DeepCanopy) has a lot of potential for improving safety at job sites.”
DeepCanopy proposes to provide that safety improvement through the use of cameras equipped with artificial intelligence software that can monitor construction sites and industrial environments for safety concerns.
Fuller describes the cameras as akin to extra pairs of eyeballs for supervisors monitoring job sites and maintaining safety. Instead of one person monitoring multiple video feeds, the cameras can spot issues on their own and notify the supervisor via email, text message or something more direct, like a siren.
“We’re not looking to replace existing safety practices,” Fuller says. “The core value proposition of the product is that it’s increasing safety at job sites.”
One of the biggest things the cameras are looking for are unsafe interactions, Fuller says. That could be something as simple as a worker being in the wrong place at the wrong time when a piece of heavy equipment is in use nearby — an easy-to-make error with enormous potential consequences.
The company started in Woodland, incorporating in December 2017. In September, it moved into a new office in downtown Vancouver.
“That’s when we really started building up the team,” Fuller says.
The company has grown to five employees — three at the main office and two who work remotely. The staff is a mix of software developers and industry professionals. The new office is still quite small — just an enclosed room at the CoLab co-working center in downtown Vancouver, shared by Fuller, Parnell and computer vision engineer Julian Weisbord. But Fuller says they’ll be eyeing another move as their staff continues to grow.
“This is a stepping stone toward bigger offices,” he says.
Fuller says most of DeepCanopy’s target customers are large construction and industrial companies. Ideally, he says, an employee overseeing safety would take the lead in acquiring and setting up the technology.
DeepCanopy found clients within its first six months and is in talks with two companies about deploying the technology, Fuller said, declining to name them.
Fuller’s background is in engineering and he said many members of his own family are involved in trades work, which makes him appreciate the kinds of safety improvements that DeepCanopy can make. There’s an unmet need for safety monitoring equipment in industrial industries, he says.
Since the cameras are intended to be deployed in rapidly shifting industrial areas and construction sites, Fuller says it was important to design the system to be as self-contained as possible. The cameras operate individually and connect wirelessly using cellular connections, so all they need to function is a power supply. They begin monitoring as soon as they’re switched on.
The artificial intelligence portion of the job is cloud-based, Fuller says, meaning the camera feeds are sent to a more powerful computer to perform the safety analysis. But Fuller said the company is looking into enabling the analysis to be performed on the camera hardware.
The initial focus is safety. But Fuller says the team is also developing productivity applications that will allow the cameras to recognize things like “time on tools,” meaning the amount of time that pieces of equipment are in active use as opposed to simply being present on a job site.
In the midst of a booming construction industry, he says, companies are eager to find ways to save on costs. A time-on-tools measurement could help them build more efficient inventory systems.
DeepCanopy’s development team is focused on the recognition software rather than the camera hardware, Fuller says. The hard part is training cameras to recognize unsafe practices in an industrial environment.
The process of “training” the software involves the use of large data sets from job sites, with input from industry experts to make sure cameras can recognize not only specific objects such as vehicles, but every type of equipment brand at a job site.
Fuller describes DeepCanopy’s software as robust enough to take a one-size-fits-all approach — it requires little to no customization and can be quickly deployed at any site.
The idea of always-on monitoring cameras and AI recognition software has led to concerns about privacy and monitoring. But in the case of DeepCanopy, Fuller says the product is specific enough to mitigate those concerns, as well as concerns about the potential for abuse of the technology.
The software is trained to identify things that are relevant to construction and trades industries, Fuller says, which means it can’t simply be reconfigured to search for other things, because it wouldn’t know how to spot them.
“It’s not magic — it takes time and effort to provide context for what you’re looking at,” he says. “The technology is very specific to what you’re building it for.”
