GPU Startup Story: Raytrix Conjures 3D From Single-Lens Camera

by Gary Rainville

[Editor’s note: Nearly three dozen companies participated in the Emerging Companies Summit, held during NVIDIA’s GPU Technology Conference earlier this year. Below is one in a series of company profiles showcasing how startups are innovating with GPU technology.]

When the shutter clicks on a Raytrix 3D light-field camera, you don’t take a single shot – you take every shot.

That’s because the German startup places an array of microlenses – up to 20,000 of them – between a normal camera lens and the camera’s image sensor. It’s like having 20,000 little cameras clustered together taking shots of a scene from slightly different angles.

Raytrix R11 3D Camera
Raytrix offers state-of-the-art light-field cameras with incredibly high resolution.

Instead of relying on the single point of focus of a single lens, the microlens array records every  focal point through the entire depth of a scene. By tying all that photographic data together – processed using GPU technology – Raytrix cameras record every possible image within the field of view.

To illustrate this, imagine a boxer in a ring, his trainers in a corner and spectators behind them.

By capturing this scene’s extended depth of field, Raytrix cameras can focus the resulting image on any vertical plane – from the tip of the boxer’s glove, to the towel in the cornerman’s hand in the middle distance, to a cheering spectator in the very back of the arena – all from a single shot.

Raytrix cameras can even focus on particular pixels and view images from slightly different angles. Plus, they work with everything from microscope to telephoto lenses.

Raytrix software, developed using the NVIDIA CUDA programming model, processes the data from each microlens to reconstruct images according to the user’s choice. Doing this in real time requires the immense processing power of NVIDIA GPUs, which can process 6-11 frames a second, working with an 11 megapixel image. A CPU, by comparison, would take anywhere from a couple seconds up to a minute to process a single frame.

Raytrix’s products are primarily focused on industrial markets and for research. Using USB or similar cables, they attach to a PC outfitted with an NVIDIA GeForce GPU for heavy-duty yet cost-efficient processing. The benefits to fields such as facial recognition, fluid dynamics and microscopy are enormous.

Take automated quality assurance. An automobile part, for example, coming down an assembly line can have different heights and various points that need to be inspected. With 2D, either the product or the camera has to be moved to examine these different spots. Some faults, like a crack in a bolt, would have to be estimated – rather than actually seen – because it may not fall into the limited focus point.

In contrast, a Raytrix camera can take a single fixed image, which can then be scrutinized from front to back, at differing angles and in incredibly high resolution 3D. It’s faster, more cost-effective and more efficient, which can be particularly useful to round-the-clock operations.

The 3D light-field photography market is only in its beginnings. Christian Perwass, co-founder and general manager of Raytrix, is already looking to the future, however. 3D video is the latest hurdle his company has crossed, which could be a boon to the 3D television and film markets.

Next on his wishlist is building the processing into the camera so it could work with any computer, and minimize code changes. In the long term, shrinking power constraints and expanding processing power could one day allow the technology onto mobile devices – and enable many more uses.