Motion Capture - Computer Vision for Visual Effects - page 258

(a) (b)

Figure 7.3. (a) A calibration device used to precisely spatially calibrate cameras for motion cap-
ture. (b) An image of the calibration device (circled) from one of the infrared cameras, showing
the highly retro-reflective markers. The other bright spots in the image are the infrared strobe
lights from different cameras in the capture volume.

The camera system is precisely spatially calibrated before each capture session
using a special device, as illustrated in Figure 7.3 a. This device is usually a rigid
wand with several markers at measured intervals along its length. As the wand is
moved through the capture volume and observed by the surrounding cameras, it
generates feature matches across all the images. As illustrated in Figure 7.3 b, finding
and tracking the markers on the device is an easy image processing problem since
under infrared light they appear as bright dots on a dark background. Therefore, in
each camera at each point in time, we can uniquely identify several image feature
points (which can be disambiguated due to the uneven spacing of the markers).
Collecting all the device observations from each camera together gives a set of image
feature matches where i ranges over the cameras and j ranges over the

unknown 3Ddevicemarker locations. This collection of featurematches provides the
input for amulticamera calibrationproblemthat canbe solved exactly as described in
Section 6.5 . The known physical distance between the markers is used to recover the
scale ambiguity resulting from Euclidean reconstruction. The reconstruction errors
obtained with this controlled procedure are very low— less than a millimeter.
The performer wears a tightly fitting body suit with spherical markers carefully
attached near joints of his or her body. The markers range from five to thirty mm in
diameter, and are retro-reflective , meaning that they strongly reflect light rays along
the vector pointing directly back at the light source . 3 T herefore, the markers are easy
to distinguish in each camera's infrared image, since they appear to be extremely
bright, as illustrated in Figure 7.4 . The retro-reflectivity is also important since each
camera has little time to gather light from the scene due to the very high frame rate
of motion capture systems (e.g., 120 Hz) required to capture fast motion . 4

{ ( x ij , y ij ) }

3 This phenomenon is similar to a bicycle reflector or a cat's eye, and is typified by 3M's Scotchlite
material.
4 An alternate optical approach gaining in popularity is the use of active-lighting markers, such
as small red LEDs that encode a unique identifier by blinking at high speeds, as in the systems
produced by PhaseSpace, Inc.