3d Tracking (Match Move) ....A Doubt

[plainman007]

Hi, I finally worked out how to track within 3ds max. Tedious Process.

My question is that ive learnt how to track markers so the 3d cam replicates the exact same movements of the real world cam that took the footage. But how can i get the camera to fly in from far away ? How do they do that in the movies. They show u a guy standing in a balcony of a hotel like from almost one mile away and the cam comes flying in and slows down near the actor and then orbits around him. Im able to do the orbiting around but i obviously cant put my camera on a one mile long jib arm ?? So how does one makes CG scenes which are much larger in scale to the actual chroma tracked set ? How do we get distance ? Ive had some ideas of having a second cam in max and then animating a flyaround that shows many buildings etc crossing by then comes slowly closer to the matched cam and ends on almost the same position where the tracked camera starts. So i can continue the shot from there using the matchmoved camera. Now ill render out 2 segments and join them in post so that i get one long seq of the flyin. BUT ... how will the greescreen footage match to the non matched 1st seq cam ?

I hope u get what i mean ....Id be glad if someone could help.

Thanks
Plainman

[Anhar Miah]

Well, I too have doubts myself about doing these kinds of shots, and have come to pretty much the same conclusion:


Segment (a)

(1) Shoot Character in greenscreen, (have a X duration static camera shoot for later use) with tracking markers.
(2) 3d Matchmove the shot and export 3d data into 3d application
(3) Create 3d environement
(4) Composite 3d background with greenscreen footage

Segment (b)

(5) Using the same 3d environment extent out to allow for 1 mile fly in, render this section AND make sure that the last Frame is the same camera postion/angle as the first frame of the segment (a)

(6) fire up the compositor and use the X duration static shoot into the "Flyin", basically as the camera "flys in" your character is gonna be "Zoomed in", so you can start off with him scaled down to a tiny dot then as the camera "flys in" you slowly scale him up at the same time (giving the illusion that we are actually getting nearer to him) also you may want to make any other translaction such as up/down if the camera moves in that way. BUT what you can't have is ANY rotations since the X duration shot is from a fixed perpspective.

I hope some of that made sense, but that is how I would do it.

Anhar

[NickJushchyshyn]

There a lots of approaches.
If you're extending the camera move, you can do it in 3D by just adding key frames to the camera created by the matchmove process. You don't need extra cameras.
If you pickup Tim Dobberts matchmoving book, I think there's a whole chapter on extending camera moves.

There's an example of a shot like this for the opening shot of this little fx experiment:
http://www.pixelcorps.com/Big_Freeze.mov
Was mostly an experiment and quick turnaround, so it's nothing amazing, but it shows the concept.

Just the end of the move was matchmoved. (Photo from the shoot is shown in this article from millimeter: http://digitalcontentproducer.com/dc...cts/index.html)
Up to that point, the people (shot on greenscreen of course) are actually just a single frame from the greenscreen plate placed onto a 2D plane.

Any specific questions?
Have fun.

[plainman007]

Anhar > Yes i figured it would have to be that way but as i thought its a pseudo match move done by hand. More like a cheat. Not that theres anything wrong with that as long as it looks convincing. Actually even if you nottice most of hollywood movies where they fly from around a city and into one window. It looks so obvious that the scene inside the window is lacking depth. Then as the camera move into the window you will most certainly see a blend as if there was a fade from the 2d video into a more dimensional video. I guess its because of this trick. I figured theres a way to take this a step further to make it more dimensional. We will have to shoot greenscreen footage with an intial leader fill in time with an approximate of the angle from which the flying in cam will be coming. (needs adv planning as to which direction the cam will come in from then the camera man would mimic the anlges alone very slowly giving time because the cam is flying inwards (zooming) while slightly changing perspective and angles alone) Now we would use this advance lead footage as an avi with opacity mapping onto a plane with a look at constraint applied to it. Meaning the plane always faces the 3d cam. So as the cam comes in and finally reaches the last frame (next frame onwards is matchmoved) we will delete the plane and render the rest of the scene as is. So then use compositing to match the rest of the matchmoved chroma footage. The advantage here is that the people in the chroma can be waving or moving etc instead of a still frame. So in short your real camera would have shot exactly 5 secs of lead footage if your flying (untracked) cam is 5 secs long. This should work as far as it goes ..pardon the pun.

Nick > Thanks for your help...Im downloading your video as im typing this. When hollywood makes building flyins etc. Do they actually make the entire 50 to 60 buildings or is there any sorta layering involved. Like lets says they keep flying in the camera but keep adding buildings and deleting ones that have crossed the camera (to keep geometry low). Or do they realy make that many boxes and map them all with windows etc...huh must be tedious then ? And in CG scenes like that ive noticed after a couple of buildings they fade and the rest looks like a matte painting. Do they do that. Use a certain number of boxes and matte paint the rest ? If your answer is yes. Then how does the matte move in perspective compared to the cam moves and other buildings ?

Thanks

[NickJushchyshyn]

Basically, the answer to all your questions is .... "yes".

In the move in our video, everything behind the cable car is a flat matte painting (actually four of them). Anything that the camera doesn't move close to can usually be done with a matte painting.

Mid range stuff can often be done with simple cubes that have detailed building textures painted onto them.

The stuff really close to the camera usually needs to be full 3D.
In this case, individual buildings (or small groups of buildings) can be rendered seperately (usually with seperate renders for color, reflection, specularity, even distance from camera) then everything gets comped in 2D.