The latest safety-savvy vehicles may have as many as eight to ten embedded computer vision cameras in them, mostly to assist with autonomous driving. The end game is ultimately safety, but with these new innovations can come danger, as well.
In extremely cold or hot temperatures, the lenses in a camera embedded into a vehicle’s exterior can contract or expand, which can lead to blur. And if that blur leads to a stop sign or a pedestrian being misrepresented as something entirely different, fatal consequences can happen.
This is just one of the many real world applications that Montreal startup Algolux and it’s band of PhD researchers in computational imaging are solving.
The company, led by CEO Allan Benchetrit, is reinventing image processing for next-generation cameras. These can be smart phone cameras, DLSRs, security cameras, medical imaging cameras and more. Rather than choose to attack one problem, the group of 16 employees has actually built out a software-based image processing platform, called CRISP (Computationally Reconfigurable Image Signal Platform), that can allow them to rapidly prototype and bring solutions to market for any number of problems.
“Something that would take traditional approaches two years is taking us six to eight weeks to prototype because of this proprietary framework we’ve build for processing images,” Benchetrit told MTLinTech from Algolux’s Point-Sainte-Charles-based offices.
Algolux’s approach is so unique because it’s using software to mimic and improve processes that have traditionally been hardware-based. It means Algolux can solve problems within image processing that would be very difficult with hardware, and this can lead to higher-quality images.
“Cameras are going through a revolution right now in terms of their capabilities, like innovating on optical elements of the camera, on the sensors (what can be detected by the camera) on low-light image capture, on deblurring and more. Those are things that require a lot of computation in the camera in order for them to work,” said Benchetrit. “We have the ability to do computation on the fly, as required by all of these innovations that are coming to market.”
Driven by novel sensors and complex imaging algorithms, tomorrow’s cameras will have increasingly varied configurations. This will make it difficult for hardware-based designs (highly optimized for power, speed and cost) to keep up with the pace of innovation. But CRISP, a software system that is optimized holistically for final image quality, means that errors are minimized.
Nearly all of Algolux’s 15 employees are researchers, and many hold PhDs, who review scientific literature and industry innovations to work within CRISP to create innovative solutions. Essentially, said Benchetrit, these are all people who work within applied math, like computational imaging, physics, computer science and more.
And just like in computer programming where everything can be reduced to 1’s and 0’s, each pixel in an image has a colour assignment, which is basically a string of numbers.
“A lot of what we’re doing is assigning numbers based on how we understand a picture should look,” said Benchetrit. “But the fact now is that there are more and more problems that can be solved with this in the real world, and we’re one of the few companies that is actually focused on this.”
So what are some real-world problems that Algolux can solve?
Well, there are many. Algolux has developed prototypes to develop thinner and lighter camera modules at lower manufacturing costs, correcting lens blur and problems associated with “low-light noise.” Low-light denoising is particularly important in today’s market, as many social media pictures are taken in low-light conditions.
Algolux’s work within making the lens in a smartphone camera thinner is fascinating. Their technology can actually replace the lenses in the camera with software by mimicking it. Lenses in a camera remove blur or aberrations that are caused by light refraction.
“Our software compensates for those problems algorithmically. You can replace the work of those lenses with, essentially, math,” said Benchetrit.
“There’s a lot of cameras and new uses coming to market, and we’re doing significant research down the line,” added Algolux’s director of marketing, Daniel Nahmias-Léonard. “As new cameras, sensors, functions and applications come to market, we’re going to be one step ahead of those trying to play catch-up.”
As for potential clients for the startup that raised $2.6 million from Real Ventures in 2014, Benchetrit said it could be anyone from smartphone manufacturers, to the Intels and Qualcomms of the world that could license Algolux’s work to replace or compliment the ISP chip (image processor) they currently use. Or, it could even be camera sensor manufacturers, like Sony, Omnivision, and more.
“The plan now is to continue to come up with these prototypes, collaborate with very large companies and hopefully that leads to commercial agreements,” said Benchetrit. He’ll also look to add more world-class scientists to his research team before the end of 2016.
Along the same timeline, Algolux will also seek to raise a new round of funding, likely in the $1-2 million range, to evolve the CRISP platform and help assist in negotiation efforts towards commercialization.