Camera Calibration for 3D Reconstruction and View Transformation - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 112

they can be used directly for other applications, as well, such as body, face, and
gesture modeling and animation.

3-D Face Model Reconstruction

This application attempts to compute the facial action parameters ( FAP s)
automatically from a pair of stereo sequences of the human face. For this
purpose, the face is marked with blue dots (see Figure 2).

Manipulation process

The whole process includes five stages and can be briefly described as follows:

1) Calibration : The employed convergent stereo set-up can be calibrated

using any method discussed in the last section. However, since the
calibration results are only used for 3-D reconstruction, an implicit
calibration together with the use of the reconstruction-distortion model
would be sufficient. The distortion coefficients are calculated by the plumb-
line method (Van Den Eelaart & Hendriks, 1999). After the distortion is
removed, the linear DLT method is employed to recover all DLT param-
eters encoded in ~ .

2) Recording : Without changing the pose or any internal parameters of the

two cameras, a person with some special (shape, color, and position)
markers on the face sits in front of the cameras, in the common view volume

Figure 2. One example face image. Left: The original image. Middle: The
image after a hard thresholding. Right: Markers to be tracked through the
sequence.