Barren landscape 3D camera track woes

[Blaise Douros]

Precompose, and in the precomp, sharpen, add stupidly high amounts of contrast, darken it, whatever it takes to make the features more visible, and track the resulting precomp with AE’s 3D Camera Tracker. I’ve done this a number of times. Also, if you have the metadata about the shot (sensor size, lens focal length), do a calculation to obtain the horizontal Angle Of View, and manually enter that into the 3D tracker. Use the Detailed Analysis checkbox. All these things in combination may result in better outcomes for you.

[Richard Garabedain]

Shots like this are harder to track due to their low angle and the large flat plane. Even in mocha these shots are hard to get right…but what if you only used the tracker to create a camera..then set up your graphics on a plane with shadows and then just winged it with your eyes

[Boris Dimitrov]

Those are excellent tips! I tried it out, and it really seems that the texture & contrast comes out sufficiently, but the AE tracker still fails. Here are 2 frames from the modified clip, 3-4 seconds apart. I uploaded two frames just to show how drastic the change in perspective changes the features on the ground, I am suspecting that this might be the reason why the track is still failing. I’ll keep experimenting to see if I can make it work with different contrast/sharpness and other image modifications.

I did have some success with Mocha Pro, managing to decently solve for a segment of about 1/3 of the total clip. I did it using:
– a unlinked (“Link to track: None”) layer of the foreground (ground closest to the camera) that covers 1/2 of the width of the frame in the center. I meticulously adjusted the surface plane to match the perspective of the ground plane as close as possible. Man it takes a long time to get it to feel right, and I am never quite sure (there are zero references to the perspective line).
– a second unlinked middle ground layer, covering the ground plane further in the distance, covering about 1/3 of the frame width in the center of the frame, once again with a meticulously adjusted surface plane.
– a third (linked) layer that covers the mountains in the back with a perfectly vertical surface plane

Using those 3 layers, the camera solved with 96% accuracy and had a decent track in AE. Problem is, even though those layers’ splines are fixed and do not move with the track, their surface planes/grids do and once they go near and beyond the edge of the screen, they warp in strange ways. So I could only get a good camera solve for a segment of the clip where the planes were mostly within the clip.

Last night, I tried making a second set of the foreground & middleground layers that start at about the place where the first set run off screen (setting out points for the first set at around the same place), but that worsened the accuracy of the camera solve to 92% and produced an unworkable track when brought into AE. Ill try and refine the second set to try and make it work.

I’ll keep experimenting. Ill probably buy a few planks, mark them with “X”-es and spread them out at different Z-depths in the actual shoot with the hope of being able to make it work.

[Boris Dimitrov]

Thanks for the suggestion! I am not 100% sure what you mean – use a track that gets poor results in terms of tracking points, but that can be used to create a camera, then manually place the objects in 3D space (not using the tracking points for solids/nulls)?

If the track is poor, wouldn’t there be a big risk that the camera movement is also off? I am concerned that I might end up spending a lot of time trying to get the composited layers to not float when it turns out that that was impossible due to poor camera movement data.

[Richard Garabedain]

yeah…there is a risk the camera is not right….Mocha has a way to track planes for such an occasion..but it takes a tutorial to learn

[Blaise Douros]

Post a Dropbox copy of the clip? I’d be curious to try–I’ve rarely come across a scene like this that AE can’t solve. Sometimes you have to mask out part of the scene to limit AE’s confusion.

AE looks, if I remember correctly, more at the light areas than the dark, so if you’ve got good contrast in the ground, you’ll be better off.

One other trick that has worked for me in the past is precomposing and reversing the clip. Weirdly, AE tracked that better than it did forward. Then you just time-reverse the keyframes of the camera.

[Boris Dimitrov]

Hi Braise,

Sorry, been out and away from my computer for a while. Here is a link to a copy of the clip. I will be hugely grateful if you can crack it and share insight on how you managed it!

[Blaise Douros]

Took about eight minutes. Precomped it, and lightened the color up, increased contrast so that the plain had lots of easily-trackable black spots. Used 3D Camera Track, and used Detailed Analysis, selected Typical from the scene type, and that was it.

Here’s the AEP file: 13158_dimitrovlandscapetrack.aep.zip

[Steve Bentley]

Instead of increasing contrast etc which can add or enhance video artifacts, you can also use one of the color records to do the track. Red is usually better than green which is usually better than blue. Or you can combine different records in unusual ways (channel math) to enhance or create difference channels and track those. Its all still the orginal data without squashing the colorspace.