Skip to main content
Participating Frequently
April 10, 2019
質問

focusing 3d camera tracker on a region of your footage

  • April 10, 2019
  • 返信数 2.
  • 1232 ビュー

It seems like it would save resources if you were able to 3d camera track specific regions of your footage. Can this be done using masks? Or do you have to track the whole scene then eliminate track points after the fact?

    このトピックへの返信は締め切られました。

    返信数 2

    Community Expert
    April 10, 2019

    What, exactly, are you trying to do?

    I did a project once where I used Mocha AE to corner pin track the left cheek of an actor, expanded the surface, applied the corner pin track to a solid, applied a grid to the solid, pre-composed just the grid layer, camera tracked the Grid, established an origin and ground plane, added a solid and a camera, used the parenting trick to move both the solid and the camera so the solid was at comp center, exported the camera and solid as a C4D file, added a 3D model I created in Blender of really badly burned face with the jaw bone exposed, then imported that into AE and gave my actor a truly disgusting left side of his face that was a completely believable 3D overlay on the face.

    Let us know exactly what you are trying to achieve and there is probably a solution.  If it is nothing more than surface replacement then the basic Camera Tracker is or Mocha AE, or even corner pin tracking is probably what you want to use. If it is more complex then the existing tools can probably be used in a more creative way to pull off the shot you are trying to create.

    Participating Frequently
    April 12, 2019

    So, i work with drone footage of golf courses. Placing text on the ground plane is often essential but when all of the ground is the same color it is frequently overlooked by the 3d camera tracker. Ill get track points on the trees and ther signifigant objects but very few or no track points on the ground.

    Sent from my iPhone

    Community Expert
    April 12, 2019

    Try this. Duplicate your footage, add some significant color correction to the duplicate to bring out as much contrast and detail as you can and don't worry about how it looks (I recently applied Colorama to a scene to get a decent track and it was all bright green, orange, and blue, but at least I had some detail to track) Pre-compose the color corrected footage, Camera track the Pre-comp, Make sure you establish an origin and ground plane then add a camera and Solid using the same points, add a Grid to the solid to make sure that the track is good, then start selecting track points for the areas you need to add graphics to and try and keep your points as far apart as you can to make sure the surface track is accurate. I usually put solids everywhere I want to add something, add a grid and make sure the track is good, then I add my graphics or replacement layers, make them 3D and parent them to the reference solids while holding down the shift key. This snapps the replacement layer to the same position and orientation as the reference solid. I then just turn off the solids or delete them.

    Once you have your 3D scene set up you can just delete the color corrected pre-comp you used to get a good camera track.

    BTW, is it also a really good idea to make sure you have removed as much lens distortion as you can before you camera track. AE's camera tracker does not like barrel distortion in lenses at all.

    If you can't get camera tracker to work properly you can still use Mocha AE, Corner Pin track or even Motion Stabilize just about any area of the golf course and then insert your graphics and put the motion back in the shot. I've done that kind of trick many many times on shots that just did not lend themselves to camera tracking.

    Mylenium
    Legend
    April 10, 2019

    AE's implementation is fully automatic, so you can't manipulate the track marker placement beforehand.

    It seems like it would save resources if you were able to 3d camera track specific regions of your footage.

    The whole point of a 3D tracker is to determine an actual camera placment, so in order to determine sufficient parallax, the points have to be spread far and wide and there have to be quite a few of them. Therefore masking out entire areas doesn't even make sense - if you were to limit the tracking to a cropped region, the solution would be very likely extremely off since the solver doesn't have enough info to reconstruct things like lens curvature and relative movement of the tracked points based thereon. Masking and cropping before a track therefore an absolute no-no. Of course manually placing and eliminating autoimatically generated track points after the initial analysis is a different story, but if you need that, you have to look at tools like SynthEyes. The rest is mostly a moot point - most 3D track solvers don't care for the actual number of points. That's simply not how this stuff works. Forcibly reducing the number of tracked features would be more a matter of the human side of things, but not a way to optimize the tracker. Even AE's limited tracker can handle a few hundred points without issues and most commercial alternatives can track humungously large scenes with potentially millions of track points.

    Mylenium