Is this possible?

[hype]

I'm just throwing this out there, I have no clue if this is possible or not, but it seems to me like it should be.

doing a four corner track in AE is a great way to cheat a 3d track (like boujou or syntheyes), as long as the image you're tracking in does not go beyond the boundaries of the rectangle made by the four corners you tracked.
If the image you want to track in does extend outside of that tracked rectangle, what do you do?

You can't parent to the corner pinned layer, because corner pin is an effect, not transform data.
will the Grow Bounds effect manage this somehow?

Shouldn't there be a way to calculate a 3d camera's position based on the relation of each animated corner point to a fixed point in space? after calculating the differences, that fixed point would then use the data as "weights" (to use a 3d term), telling it which way to angle and pan, thus becoming the animated camera. The corner points could be "attractors", pulling the camera in that direction more if it strays more from the starting point (or whatever control point it would use).

Am I crazy? Should I stop thinking about this? Or does this sound like something a simple script could do, make a little animated node out in space that you could parent a 3d camera to?

[Mylenium]

Why go the long way round? Revision FX' Re:Map plugin set is all you need. Their "Planar" allows dilation, skewing and perspective correction.

Mylenium

[hype]

Hi Mylenium,

Thanks for pointing me to that one! But, it looks like that one still doesn't let you pin an image that exceeds the space of the rectangle made by the 4 corners. Does it? It doesn't list anything like that in its feature description.

I just can't help but think doing a simple planar 3d track in AE would be super easy with a few expression or a script. We all know it perfectly handles 3d and has a great 3d camera and environment, and the corner pin tracking gives you all the data you need to move the camera around. Its just a matter of transfering that data from the 4 points to the camera pan and tilt.

[vidpat]

Unfortunately, this is not quite as easy as it sounds. If you go through the mathematics of it, the problem is hidden somewhere in there: For any four points in the image plane, there are at least two solutions for the distance of each of those four points along the vector to the camera, thus several possible 3D planes that could map to those points. In this case, it may be difficult to predict the position of the fourth point because of that.

I was working on something similar to this problem a while back, but haven't had a chance to revisit my code in a long time.