Quote: "The maths required in camera tracking are a little more complex than you would think . Ive looked into it."
I'm sure their system is very complex! But my initial thoughts are a MUCH simplified system.
-You have 3 coloured balls, in primary colours.
-You put them anywhere in your scene, then measure exactly how far they are apart on all 3 axis (get your tape measure out!). You effectively give them 3D coordinates.
-You store these in your DBP program and you now have a 3D triangle that represents your base orientation
-You now code an image processor that looks for these primary colours in each frame of footage, and figures out where the center of each ball is in the frame (obviously the balls will always have to be in shot and you'll need to work hard on this code to make it as reliable as possible)
-You can now calculate the new triangle.
-You always assume you've above the triangle on the y, so you can get exact rotations and use EZRotate (to make life easier!) to calculate the offset of rotations the shot has to the base rotations, as well as the zoom and position (I personally think this would be relatively straight forward)
-Position the camera in 3D space accordingly
Initial problems I see:
- Accurate recognition of the balls
- Having 3 coloured balls in your scene! You'd have to put 3D objects over the top of them in your compositing
- Error handling and smoothing of incorrect angle calculations etc.
I'm sure it'd be tricky and fiddly, but I dont think it's that complicated. I think the genius in their software would be the reliability, and their much more complex cue recognition. Still, I reckon it's doable. The main thing that'd stop me having a try is I can't see a real use for it except for messing around, since DBP isn't geared towards photo-realistic rendering.