Facebook’s open source Surround 360 captures 3D-360 video at up to 8K
Facebook has introduced the Surround 360, which captures 3D, 360 degree video using a total of 17 cameras and can output resolutions of up to 8K per eye. Unusually, Facebook will be making both the camera and processing software open source to give developers the opportunity to improve both.
The Surround 360 itself features 17 synchronized cameras: 14 horizontal, a fisheye on top and two more on the bottom. Each camera has a global shutter (which eliminates rolling shutter) and has been designed for long periods of operation without overheating. Raw Bayer data is captured, which is later processed in the stitching software. All 17 cameras are bolted onto an aluminum chassis so everything stays in place.
Facebook says it has used Point Grey industrial cameras in the Surround 360, which hints at the use of Sony 2nd generation Pregius CMOS sensors with global shutters. The lenses used are 7mm F2.4 lenses designed for up to 1″-type sensors, which could even mean the use of the latest Sony IMX253 or IMX255 chips. If that’s the case, then these lenses are roughly equivalent to 19mm.
With incredible amounts of data coming from all of those cameras Facebook uses a Linux-based PC with a RAID 5 SSD array that shares the writing out across eight drives simultaneously. The company has made controlling the camera rig easy, via a web-based interface that allows users to adjust shutter speed, exposure, frame rate and gain.
The stitching software uses the concept of optical flow to resolve disparities between what pairs of cameras can see. The company says this method is ‘mathematically trickier’ than traditional systems, but yields better results. The end results are 3D/360 videos which can be output at 4K, 6K or 8K per eye. Videos can be viewed on Oculus Rift and Gear VR headsets using Facebook’s Dynamic Streaming codec. Videos can also be output and shared on Facebook and other websites.
A big part of the Surround 360’s story is that Facebook is opening up both the camera blueprints and processing software to developers, stating that ‘we know there are ideas we haven’t explored’ and ‘we know from experience that a broader community can move things forward faster than we can.’ The company says that the design and code will be on GitHub this summer.