3D vs PoV

So this is a short (but hopefully helpful … because who doesn’t make stereoscopic films on a daily basis) discussion of a couple of technical consideration with stereoscopic (3D) footage.

Basically they all revolve around maintaining persistence of vision (PoV) when viewing stereoscopic (3D) footage. Commonly associated with animation, PoV is actually universal to film (and any motion graphics). PoV is the phenomenon where our brains ‘believe’ that a sequence of images are acutally moving. Basically with 3D footage our minds become very susceptible to a losing this ‘belief’. Poor fragile squiggy things.

Camera Motion

The first thing to get in the way of PoV is in unfortunately one of the cornerstones of cinematic craftsmanship - camera motion. Complex and erratic camera moves cause our minds to question the validity of the PoV of the footage that we are watching. Hand-held camera moves (particularly if the operator turns) is a real issue. Linear tracks or zooms don’t suffer from this nearly as much, perhaps due to a reduced inconsistency between inner ear motion detection and the visual feedback. Likewise a slow and steady pan or tilt allows for the maintainence of PoV. This issue is dramatically compounded if the imagery doesn’t take up much of your field of vision (i.e. a little TV or from the back of the theatre). This is caused by our brain preferring to go with the motion that is evident in the periphery vision that correlates to what the inner ear is telling us is happening, and consequently ignores the 3D affect of the vision. The immersion of IMAX projection provides a great way to circumvent this … so make sure you have an IMAX theatre on hand when you want to impress the kids (or the ladies).

Framerate

The first thing to do to fix the above issue is to crank up the frame rate - in fact that helps with almost all of the issue mentioned here. 3D games are much more convincing than 3D films purely because they typically run at much higher frame rates. (i.e 60 fps vs 24 fps). The common misconception is the 24 fps is what you need for POV, but that isn’t true in the slightest. A lot of slow moving footage will create PoV at 12fps, while most first person shooter games look dreadful and jerky if not running at +50fps. The 24fps is a run of thumb that was applicable when the world was a whole bunch simpler. There are some theories (which I agree with) that PoV is based on the maximum change of position of a point (i.e. object) in the field of vision proportionally to the ratio of change to the overall field of view. Maybe it looks something like: Ratio (delta((xT - xT-1) + (yT - yT-1)) to (delta(sum(screenT)-sum(screenT-1))… or maybe not!

Lens phenomenon

The key lens phenomenon to mention are motion blur and depth of field (DoF). One can aid in the PoV (motion blur), the other competes with PoV. Although optically, our eyes do have DoF, we don’t perceive it as our eyes/brain continuously adjust the focus point without our conscious control. This mean the cinematic camera effect (quite often employed for artistic affect), is something that adds to the suspension of PoV. An obvious example of this it the first shot from “Avatar”, where two dots appear to be floating on the ‘zero’ plane, but they are in fact an out-of-focus bubble with significant foreground depth.

Convergent Lenses

Basically there are four stereo lens setups that can be used for 3D imagery. They all have their own positives and negatives. They are:

• Fixed Convergent (for static 3D this is probably the best solution)
• Varying Convergent (this is how our real eyes work)
• Parallel (lacks the “zero depth” plane, which nothing is appears infront of the screen)
• Offset Parallel (as above but the L/R images are moved slightly so that they lineup at a certain depth - results in a foreground/zero plane/background effect)

Convergent Lenses

Parallel Lenses (the separation is consistent with the other renders.)

Offset Parallel Lenses (without overscan)

Note this is a point of conjecture, some people believe lens convergence allows the ‘focusing’ of the viewers attention - I think Nah! - it just makes the brain’s internal 3D calibration more complex, and therefore the affect slow and easier to lose. This more of a point for purely CGI 3D, as convergent lenses are ridiculously complex to setup (and maintain) when working with real life cameras.

Also a quick note to all those 3D cinematographers out there, a larger lens separation doesn’t create a better 3D affect. And if someone tells you that the lenses should be separated by 7cm (the typical distance between human eyes), then tell them to shut up - they evidently haven’t ever made a 3D film with anything other than wide shots in it!