Proceedings Article | 8 February 2008
KEYWORDS: Visualization, Computer graphics, Digital imaging, Metrology, Time metrology, 3D image processing, Cameras, Eye, Stereoscopic cameras, Composites
A primary requirement when elements are to be combined stereoscopically, is that homologous points in each eye view
of each element have identical parallax separation at any point of interaction. If this is not done, the image parts on one
element will appear to be at a different distance from the corresponding or associated parts on the other element. This
results in a visual discontinuity that appears very unnatural. For example, if a live actor were to appear to "shake hands"
with a cartoon character, a very natural appearing juncture may appear to be the case when seen in 2-D, but their hands
may appear to miss when seen in 3-D.
Previous efforts to compensate, or correct these errors have involved painstaking time-consuming trial-and-error tests.
In the area of pure animation, efforts to make cartoon characters appear more realistic were developed. A "motion
tracking" technique was developed. This involves an actor wearing a special suit with indicator marks at various points
on their body. The actor walks through the scene, then the animator tracks the points using motion capture software.
Because live action and CG elements can interact or change at several different points and levels within a scene,
additional requirements must also be addressed. "Occlusions" occur when one object passes in front of another. A
particular tracking point may appear in one eye-view, and not the other. When Z-axis differentials are to be considered
in the live action as well as the CG elements, and both are to interact with each other, both eye-views must be tracked,
especially at points of occlusion.
A new approach would be to generate a three dimensional grid, within which the action is to take place. This grid can
be projected, onto the stage where the live action part is to take place. When differential occlusions occur, the grid may
be seen and CG elements plotted in reference to it. Because of the capability of precisely locating points in a digital
image, a pixel-accurate virtual model of both the actual and the virtual scene may be matched with extreme accuracy.
The metrology of the grid may also be easily changed at any time, not only as to the pitch of the lines, but also the
introduction of intentional distortions, such as when a forced perspective is desired.
This approach would also include using a special parallax indicator, which may be used as a physical generator, such as
a bar-generator light and actually carried in the scene. Parallax indicators can provide instantaneous "readouts" of the
parallax at any point on the animator's monitor. Customized software would equate as the cursor is moved around the
screen, the exact parallax at that indicated pixel would appear on the screen, immediately adjacent to that point.
Preferences would allow the choice of either keying the point to the left-eye image, the right-eye image, or a point midway
in-between.
