Stereo Video Production : Introduction
Scott Lawrence (formerly with Vuzix Corp.)
Producing 3D Stereo content is a confusing and complicated task. There are a lot of things you need to be aware of, more than just pointing your camera and recording. With this content, I hope to get across all of the knowledge and information you will need to get stereo filmmaking within your reach.
By the time you've gotten through this, you will know the basics of stereo filmmaking, the basics of building a 3D camera if you do not already own one, the basics of editing with stereo footage, and also how to produce media assets that will play on various 3D display technologies.
Note that these pages describe the process as of April, 2008. Since then (it is December 2009 at the writing of this note), the Vuzix eyewear have switched to a side-by-side 3D as their preferred format. Side-by-side 3D is much easier to work with, rendering much of the latter portions of this book obsolete. The 3D production side of things is still the same, however. If you have a Vuzix AV920 eyewear device, you can download a firmware update to get this new functionality. There is also an image overlay you can put on your side-by-side video to make the eyewear automatically switch into 3D mode.
As content is added to this documentation this section will explain what has been updated or corrected.
|05||SDL||2009-Dec-23||Notes about the obsolete nature of field-sequential|
|03||SDL||2008-Mar-02||Content migrated over, Movies added|
|02||SDL||2008-Jan-17||skeleton content added, galleries, CSS honed|
|01||SDL||2008-Jan-16||basic outline, layout, navigation, TOC generation|
Color Anaglyph is probably what most people think of when they think about stereoscopic 3D content. It consists of color filtering the left and right eye content with red and cyan filters. (Some variants use blue, or green filters in place of the cyan.) Since all of the colors involved are reproducible on film, only one projector is required in theatres.
It's the easiest and cheapest to broadcast and view. The problems with it are that color content looks poor at best, especially when the subject is wearing colors other than shades of yellow. It is also difficult to compress for internet distribution due to the way that the compression algorithms extract and store the color information.
Another common method to display 3D content is the use of lenticular lenses. These can be found in large-format printed advertisements, DVD box covers, and so on. This method involves the use of many thin vertical strips of various angles, with equally spaced lenticular lens strips that cover each set of image strips. One eye will see one image, while the other eye will see another image. This is often used for "action" 2D shots on baseball cards and such as well. There are some "Auto-Stereo" display monitor systems that also use this technology. In these devices, the vertical image strips are replaced with LCD display elements. This allows for motion with the 3D effect. They often require that the user be at a precise distance from the display to get the correct effect. They also can produce a visual artifact of darker vertical stripes.
Polarized 3D is one of the common methods used in current 3D projection systems. It consists of two projectors, each with a polarizer on the lens system, rotated 90 degrees from eachother. Some systems, especially the latest digital projectors, will use one projector and a spinning wheel with the two polarizers. Every other image projected goes through alternating polarizer filters.
This system requires that the viewer wears specially polarized glasses.
Polarization is also used for many "3D ready" home television sets. Alternating rows of pixels on the display are polarized in the opposite direction, and the user wears the same kind of polarized glasses. The 3D content is displayed with each eye on alternating lines.
Other systems involve a sequential method, like the filter wheel method explained above, where each eye's image is displayed, and battery powered or wired shutter glasses dim each eye in perfect synchronization with the images being displayed on the screen.
This method requires expensive glasses, and specialized hardware either within the TV itself, or as an add-on box.
There are a few 3D display manufacturers that support one of these two formats to represent the 3D image data. It consists of the left and right images occupying one video frame simultaneously. They are either compressed such that they fit over eachother, or next to eachother like an old stereo view card.
The disadvantages to this method are that you lose either horizontal or vertical resolution, depending on which one is used.
All Vuzix AV products use the "field sequential" method of storing and decoding of stereographic 3D imagery. This requires an interlaced video format. There are various ways to store and represent the 480i (480 lines, interlaced) to reproduce the 3D Stereo effect properly.
All of the Vuzix devices operate with a standard NTSC 480i video source. That is to say that they take in 640x480 interlaced, at 30 frames per second. In all cases, when 3D modes are enabled, the vertical resolution drops down to 240 pixels. In the case of the '920 models, the lines are doubled to fill in the image data. In the case of the '230 models, the display device handles 240 lines natively, so no information is dropped.
Standard NTSC DVD content is 720 x 480 interlaced video format, so it is a non-issue to generate correct-working field-sequential-based 3D from these devices. The complexity of this process occurs when you have to compress the video content down to support video storage formats and devices that require lower bitrates than what standard DVD supports.
Problems will arise when some formats, bitrates or devices will convert the video imagery into a deinterlaced format, either to reduce the bitrate, or due to device capabilities. This, of course, destroys the 3D stereo data. You have to be careful to use a format and a bitrate that will preserve this information. It also means that
The intricacies of these conversions are covered in the rest of this document.
The MPEG algorithms take an entire frame, consisting of two fields, and compress it for the MPEG video stream. On standard 2D data, there usually is not much difference between the two frames so this works out fine. That is to say that a diagonal line will more or less look like a diagonal line if you're looking at the two fields interleaved with each other.
With 3D Stereo data however, these two fields may contain more different data. This means that a diagonal line far away might look fine, whereas a diagonal line close up might have a "Venetian blind" or "comb-tooth" visual effect to it. This kind of imagery requires significantly more 'data rate' to reproduce accurately.
In a nutshell, this means that a standard 2D movie can look acceptable at lower bitrates than a 3D movie. A 3D movie at the same, low birate as a 2D movie may contain lots of visual artifacts, possibly some ghosting, and probably some color bleed from one field (one eye) to the other. This may be fine for short films, but for features, this can cause fatigue in the user.
This page is a part of the Yorgle Notebook.