Limitations

< Previous | Contents | Manuals Home | Boris FX | Next >

Limitations

It's generally informative to point out what the limitations of a technique are, no matter how great it is, so that you can use it better. We'll talk about a number of these items later as well.

You must accurately know the field of view of the camera for GeoH tracking. There are a number of tools available in SynthEyes to determine it, but you must (see the section on Determining Field of View in this manual).


There's some support in GeoH tracking for compensating for Lens panel lens distortion, but that's not really the intent. The expectation is that lens distortion has been removed before GeoH tracking, via the Image Preprocessor.

You must have a decent mesh model, one that matches what was shot pretty well. (It doesn't have to be complete or accurate in parts that aren't used for tracking; likewise if some part of the model is problematic, such as hair, you can have it not be tracked.)

Note that SynthEyes's normal tracker-based object tracking (ie Object Tracking in the SynthEyes User Manual) does not require pre-existing object models; instead the 3D coordinates of the trackers are computed as a result of object tracking. If you don't have a mesh for GeoH tracking, you will likely need to do some supervised tracking to create trackers, then do a normal object tracking solve. Just to muddy the waters a little more: tracker-based object tracking can use pre-existing models as guides, especially in difficult circumstances, and you can layer GeoH solves on top of an underlying object track, especially to add secondary animation.

Note: images with square pixels are preferable. Footage that uses the Interlaced=Yes choice will be processed somewhat slower; in general interlaced footage should be avoided for any kind of tracking due to the reduced vertical resolution.

GeoH tracking is area-based: it tracks textured interiors, not exterior edges (at present). Texturing needs to be widespread, not just isolated portions of the interior, in the same way that regular 2D trackers must be spread through a 3D volume enough to constrain the possible 3D orientations. And ideally the texture should contain both large- scale texture as well as fine detail.

Faces Unfortunately, unadorned human faces aren't particularly well suited for GeoH tracking. If you start with a face, then discount the parts that move (even if they are being tracked also), you're left with texture- and contrast-free skin tones that are dominated by lighting effects. Bummer! Think bling.

Shutter time must be short enough that interior detail isn't blurred into oblivion; similarly focus must be maintained.

Tip: Evaluate shutter time in the context of the smallest feature that must be tracked. If an object is moving at 1 foot/sec (30 cm/sec), and has text that is to be tracked that is 0.25" high (6 mm), blur must be kept under ~1/10th the feature size, ie 0.025" (0.6 mm). That means a shutter speed of 1/500th second... and plenty of light! This is physics, not a software limitation.

Reflective highlights must be avoided as much as possible, as the shading will obscure the pattern being tracked and move independently of the surface itself (future SynthEyes versions may be able to reduce the effect of highlights).

The textured interior needs to be stable, ie not be doing its own thing from frame to frame. For example, floppy clothes are problematic.


Objects must be reasonably wide (thick) in the image, so that they have an interior to track, and they must be relatively wide compared to their speed of motion. Especially, they must not rotate to rapidly, in general or compared to their thickness, around the axis perpendicular to the screen.

Regular SynthEyes supervised trackers can be used to overcome some of these issues as part of a hybrid GeoH setup. You can track anything you can see with supervised trackers (at a worst-case cost of manually keying each frame); the resulting positions can then drive the GeoH solve robustly, without reference to the imagery.

The GeoH hierarchy must be just that, a hierarchy, rather than something containing closed loops, such as both of a person's arms holding a single object. You can dodge this requirement if needed, see Advanced Scenarios.

The built-in bones deformation in SynthEyes is a simple weighted scheme only... you may be able to use more advanced techniques in your 3D application, by using the GeoH objects as parents for the bones.

©2024 Boris FX, Inc. — UNOFFICIAL — Converted from original PDF.