Click the play button to view the animation. Flash required.
The very last part of this animation is in 3D (when you see Janice and I acting suspicious in the glider cockpit). To see that last part in 3D, wear red/cyan anaglyph glasses with the red lens over the left eye during that segment.
Here’s how I try to get an “optimum” amount of stereoscopic 3D depth in every image.
Watch the animation a few times, and hopefully you will understand the simple concept.
Compose the shot by visualizing the final frame of the image, and the far left and right edges of that frame.
Locate the nearest and farthest visible points in that stereoscopic frame. This includes _any_ point that is visible, anywhere, i.e. on the ground close to the camera, in the sky, far, far away, etc. Anything that is visible has to be included… _anything_.
Imagine the frame (that same final frame of the image) being located at the same exact depth as the _nearest_ point. That frame is divided into 30 equal, horizontal segments.
The left and right camera viewpoints are horizontally displaced (shifted) exactly one of those equal segments (1/30th of the frame width) apart… end of story. It can’t get much simpler than that… but…
This is actually “reverse engineering”, but don’t worry… it is easier to do than it sounds.
You can set this up any way you want… by changing the stereo base (the distance between the left and right lenses) of the stereoscopic camera… or by moving closer or farther from the near point… or by changing the focal length and the field of view… or by excluding the most distance object, etc… however you get the job done will work…
Visualizing what is going on before you shoot the scene is the tricky part, but keep practicing, and you will eventually be able to do this very quickly. If you can come up with a visual aide or two, that helps, e.g. I have a piece of paper with a vertical black bar down the center that shows me what “1/30th of the frame width” looks like…
BTW, that final frame (the “stereo window”) can be shifted, later, forward or back in the depth of the scene, but that initial frame is what defines the total amount of depth (”stereoscopic deviation”), i.e. the distance from the nearest visible point to the farthest visible point in the entire scene.
Created with Google SketchUp.
Composited with Adobe After Effects.
For simplification, I personally think of stereoscopic deviation as a “percentage” of the image width, but I also try to remember to define the angle of the FOV, otherwise it is all meaningless, e.g. my personal optimum FOV for viewing stereoscopic images is close to 33.3 deg (images with with 3.3% stereoscopic deviation). Definitely some “1/3rd” and “1/30th” laws/rules are in action, here… no doubt in my mind.
BTW, I just found similar “laws/rules” with colors in anaglyph optimization. In the red channel of the left image (in the Channel Mixer), I prefer a mix of 33/100 of blue and 66/100 of green (Peter Wimmer uses 30/70, which not coincidentally, is the same calculations rounded off to the closest 10th… you can’t see the difference between his and mine… mine is just a mathematical calculation).
There is truly something “universal” about the “1/3rd factor” (and 1/30th factor), and the strangest thing to me about that is “three of them” cannot equal one whole part (e.g. 1 / 3 = .333, but .333 * 3 = .999, not 1)… Universal “balance” apparently is not perfectly balanced. 
Click on the images for a larger view.
These images are in 3D. To see them in 3D, wear red/cyan anaglyph glasses with the red lens over the left eye.
If anyone wants to look at the discrete left and right pairs, they are located here: Cube: L1, R1, Tube: L2, R2
The outside diameter of this long tube and the width of this perfect square are exactly the same… 1.5 meters.
The tube, however, is 305 meters long.
Both of these stereoscopic images have the exact same amount of stereoscopic deviation, which is the total amount of horizontal image (or screen) parallax, measured horizontally on the flat image planes, determined by the nearest point and the farthest point in each image.
The camera setup was identical in both scenes, with the exception of the stereo base, which was wider for the cube.
So, we have two extremely different actual subject depths (1.5 meters and 305 meters), but two identical amounts of stereoscopic deviation (close to 3.3% of the image width, or 1/30th, BTW, including the black border, in this case).
How can this be possible?
IMO, our eyes and our brains are “hard-wired” for a “certain amount” of stereoscopic depth, which I call “optimum deviation”. IOW, when we look at stereoscopic images, our eyes and brain “prefer” a certain amount of depth, regardless of the subject. If this were not true, these two images would have two extremely _different_ amounts of stereoscopic deviation… _not_ exactly the same amount… or nothing even remotely _close_ to the same amount.
BTW, not coincidentally, the old, original, “1/30 stereo base rule” produced very close to the same amount of stereoscopic deviation as in these two images.
Click here for a larger image of on the thumbnail above.
Updated and animated May 16, 2007.
Click here for a larger image of on the thumbnail above.
If the word “optimum” sounds confusing, use “maximum” instead, as my “optimum” is the same as “maximum” for human eye deviation, since it is defined by the maximum amount of depth that is in clear focus when the eyes are verged on the far point.
If you are more comfortable saying “from 3 to 4 deg”, rather than “exactly 3.3 deg”, that’s fine, as “exactly 3.3 deg” is very difficult to see and measure.
Also, you might think that the lines representing the deviation (the angle to the double image of the near points) should be parallel, and if that makes better sense thinking about this in that way, fine, as again, it could only possibly mean a difference of 3.3 to about 4 deg, which is not even precisely definable with the naked eyes. The reason I have the lines slightly converging is because even when the eyes are verged at “infinity”, they are verging on one point, not two. No matter how you look at it (literally), the angle of view to the far point is never as “much” as the angle to the double image of the near point (the “deviation”). BTW, you can easily see this difference of “closer to 4 deg than 3.3 deg”, especially when the far point gets “significantly close”, by completely relaxing your eyes, instead of converging on the far point.
As you can see in the first and last frames of the animation, the double image of the near point does _not_ follow the same angle of view as the far point, i.e. it gets relatively wider and wider than that angle of view (relative to the _size_ of the near object), as it comes in closer than .9m from the eyes, and always has close to a 3.3 deg (or 3 to 4 deg) separation. When the far point gets close to the eyes, that difference between the separation of the double image of the near points is significantly wider than the angle of view to the far point (again, relative to the _size_ of the near object).
Also, the near point can never be farther away from your face than .9m to demonstrate “optimum” deviation, no matter how far away the far point is located, i.e. .9m is the maximum distance to the near point, for “optimum” deviation.
My theory has been revisied, so some of the “numbers” in posts before this one may seem a bit confusing.
See:
http://www.puppetkites.net/blog/archives/33
Hmmm. This “whole thing” seems to be “falling right into my lap”.
Don’t you just hate it when this happens? 
120 deg of stereoscopic FOV (field of view) actually includes about 20 deg of stereoscopic peripheral vision (both eyes “at once”, not one).
So, as long as you don’t Super Glue your eyes “straight forward” and staple the back of your head to a wall (in two places), so that you are restricted to looking straight ahead, we poor, wee-little, helpless, mortal humans are left with close to 100 deg of stereoscopic vision!
I don’t have to convert degrees to percentages any more!
We have another “Bingo!”
100 deg / 30 deg = 3.3 deg. That is my definition of “Optimum Deviation”! (1/30th or 3.3%… same thing)
0 deg = zero deviation, i.e. no depth
100 deg / 15 deg = 6.6 deg., i.e. “close to fusion crash” or “stereoscopic failure” due to excessive stereoscopic deviation.
I love it when things keep getting simpler and simpler.
Things People Say