R&D: 3D

Limitations Permitted: Notes on DIY 3D stereoscopic video

Limitations Permitted is filmed in British Sign Language (BSL), which encodes meaning in three dimensions of space (as well as time). The signing is filmed in 3D (stereoscopic) video in order to allow full expression of the language.



The videos are presented for individual viewing in a hand-held viewer, which is designed to run on batteries for 8 hours. Off-the-shelf 3D head-mounted displays were either too low in resolution or too expensive. Engineering a custom solution with miniature LCD screens and a file player would have been ideal, but we needed a working prototype very quickly, so we decided to modify a pre-existing portable video playback device, mounting it behind a pair of eyepieces.  
Binocular eyepieces present a magnified image to each eye at a comfortable viewing distance, rather like a ViewMaster. The left and right images can be presented independently to each eye (on two screens, or side-by-side on a larger screen), or can be encoded as a single image on one screen. We initially considered using a pair of devices with small VGA (640 x 480) displays, but managing the synchronised playback of the files under their closed operating system was not an inviting task. We were left with the single-screen option, but sourcing a device that had a high resolution screen of the required size and long battery life was difficult. Eventually we found a PMP (personal media player) with a 4.3" 800 x 480 LCD display, with a fine-enough pixel-pitch to present a smooth and sharp image through the pair of eyepieces. With an auxiliary battery, continuous playback time exceeding 8 hours was possible.
In the viewer, the 3D information could only be encoded into a single image using the anaglyph technique (using red and cyan filters) – most LCD screens do not refresh quickly enough to work with shutterglasses and other polarisation-based techniques. However, even with high-quality filters, there was considerable ghosting and smearing of the image. The remaining option was to place scaled-down left and right images next to each other on the screen, and separating the left and right optical paths with a plastic mask. This gave each eye a lower-resolution image, 400 pixels wide, but in full colour and with no interference between the left and right eyes.

For encoding the 3D information, we initially considered using a prismatic camcorder attachment that allows left and right eye-views to be recorded to alternate fields of interlaced video. Only one camera is required, though the vertical resolution of the image is halved, and there are also limitations on the recoding format (MPEG-2 based formats such as HDV do not work). The other option was to use a pair of camcorders mounted about eyes'-width apart, running in synchrony. Research suggested that an (expensive) external device would be needed to keep the cameras in sync, or at least to indicate when they drifted out of sync. However, results of an experiment with a pair of (unmatched) miniDV camcorders, started together manually, looked promising when viewed in the prototype viewer. It became clear that the critical factor was not sub-frame synchrony between the cameras, but lens alignment and matching of perspectives. When a neighbour donated an old camcorder to us, we had access to two camcorders with identical lenses (Sony PD-100 and an TRV-900). These were mounted parallel and as close as possible on a bar, levelled with a spirit level, with the lenses zoomed to the wide stop. (The camcorders' width meant that the lens separation was a little greater than average human eye separation, which exaggerated the stereoscopic effect slightly.) Both camcorder transports could be controlled with a single remote control – giving us an inexpensive but very effective standard-definition 3D shooting rig.

[written by Mukul 2009/06/07]