2r1y—seeing and measuring in 3D what you see

Ukrainian startup has achieved a breakthrough in scanning, imaging, and measuring from off-the-shelf parts

By Jon Peddie

2r1y, a startup with talent from Silicon Valley and Ukraine, was founded in 2012 by a group led by Richard Neumann and Serge Yefimov. They have spent the last couple of years building a technol­ogy for a perceptual imaging solution. Think of this as the technology that would be required to enable a computer system to see and perceive the world around it. This type of vision system will be critical for robotics, remote med­ical procedures, global security, smarter homes, consumer devices, and many other areas where machine sight and vision are important.

The directed high-speed light source can be used to illuminate a scene, or any portions or spot in a scene, and also generate a point-cloud depth map. (Source: 2r1y)
The directed high-speed light source can be used to illuminate a scene, or any portions or spot in a scene, and also generate a point-cloud depth map. (Source: 2r1y)

The problem, however, is big, too big. Sensing all the things around us with high resolution is more than big data, it is gigantic, unwieldy data. So the developers set their original goal for the design to find a means by which a finite amount of light can be most ef­fectively utilized to illuminate a subject. The result was a technology that pre­cisely and dynamically controls how light is placed into an environment.

2r1y thought the market needed a different kind of illumination (than what was currently being used) for 2D gesture-recognition systems, which could be used for smart TVs and gam­ing to capture human motion in HD at 30 fps. However, flooding a subject with kilowatts of visible light is not a desirable user experience, nor is it an energy-efficient alternative. Using a sim­ilar amount of non-visible near-infrared (NIR) light has the added factor that in more than very small amounts, a few milliwatts, NIR can be extremely dan­gerous to humans.

Originally the concept seemed sim­ple: figure out a way to provide a rather coarse spot or segmented NIR illumina­tion to enhance the performance of 2D gesture-recognition systems. The team built a test system with three banks of 100 LEDs on curved back planes. They were used to prove the idea worked, but more accuracy would be needed to make the system viable. That test sys­tem now sits on a shelf in the lab at 2r1y—proof of concept, but not prac­tical.

The company next sought to find a way to precisely place light only where needed—on a person, a hand, or just a fingertip. This allows for an extremely small amount of light to be used in the most effective way possible. The added benefit is that the total amount of energy required is a fraction of other meth­ods of illumination.

After a lot of research and experi­mentation, they found a mechanism for both controlling and directing the illumination. But then the project hit a snag when they couldn’t find a camera sensitive enough to meet their needs. Through a lot of trial and error, they evaluated several sensors and ul­timately settled on one from Omnivison.

Points of light

The core technology begins with a single light source, cre­ating a single point of illumination. At a distance of 4 meters, the point is less than 4 mm in diameter, and each point of light uses less than 88.4 nW (nano-watts)!

This is not a photograph, it is a 3D scan from 2r1y. (Source: 2r1y)
This is not a photograph, it is a 3D scan from 2r1y. (Source: 2r1y)

The system has control over the am­plitude and pulse width (time illuminat­ed) of each point. Points are generated in rapid succession and optically direct­ed over an area during a single frame exposure; think of it like a phased array radar, but with light. This effective­ly creates a dynamically configurable homogenous illumination of the sub­ject, or objects. Currently the company can get 24 million sequential points per second—that’s enough to enable 720p at 60 fps. (The team promises the next generation will have increased perfor­mance in resolution and frame rate.)

With this ultra-modulated and steer­able point light source, they could actu­ally begin to test the functionality. The device worked better than they expect­ed. They could successfully place mil­lions of points of light wherever they wanted. They began with the basics of low-level illumination and worked up to more accurate “painting” of light until they could illuminate just the fin­gers on a hand. They could illuminate a face and block light from the eyes. They could track multiple moving subjects at once and measure distances.

One night Yefimov, their chief 3D imaging technologist, suggested they see what they could do with this much control over light. Using some ideas salvaged from the original prototype, they began to play with 3D imaging. At first the results were rough, but as they worked with the system, they began to perfect the methodology and the technology. They achieved a full facial scan at 1 meter with a 1.7-mm accu­racy in 50 milliseconds—time for some champagne.

Utilizing the core module, 2r1y has begun to explore the possibilities of how to use their “directed illumina­tion.” The basic function starts with a low level of illumination, sufficient to either manually, or with AI, identify an area of interest. Like a theatrical follow spot, the illumination can then be con­centrated on a smaller area, say, a per­son. Or on smaller areas such as a hand or just a fingertip. As the subject moves or changes orientation, the illumination can follow and morph to match.

Points can be used to determine pre­cise distance measurements to a single or multiple subjects, so now they had a metrology device. Points can also be used to extract noninvasive biometrics such as pulse. The technology can also be used to generate structured light pat­terns for extracting high-density point clouds for 3D imaging.

And because every point is unique, multiple functions can be performed simultaneously (in the same frame of exposure). Current consumer-type de­vices such as structured light and pulse modulated (TOF) methods offer depth maps having resolutions in fractions of megapixels. With Kinect v2 leading the pack with 0.2 megapixels, they are not able to attain the density and speed de­manded by emerging markets, or what 2r1y has accomplished, for example, data densities 5 to 10 times greater than other systems.

CEO Neumann is quick to point out that this is their starting point. The data density is now a limitation of the resolution of the image sensor and the processing power of the GPU/CPU. You could, he states, use this same method­ology to generate a 2-, 5-, 10+ megapixel depth map. Neumann is proud of the fact they went from a white board to a functioning, market-ready camera in 88 days The company is now show­ing demos to select organizations.

What do we think?

This is breakthrough technology. If 2r1y can do what they say they can (we haven’t seen a demo yet), with COTS CE parts, this will genuinely be dis­ruptive technology. The low-power di­rected beam capability would fit into handhelds, and be useful for automo­tive applications. Auto companies are experimenting now with arrays of HD projectors on the front of a test vehicle to create shaped light around the car in front of them so the road is lit up but not the interior of the car in front. With the added benefit of being able to mea­sure distance precisely, 2r1y’s technol­ogy can be an early warning device as well. We think 2r1y has a tiger by the tail.