Biomedical Engineering Reference
In-Depth Information
modalities described in this section we will look at calibration, registration,
and the accuracy and computational efficiency with which these can be
achieved.
12.4.1
Video
Video provides real-time images of the exposed surface of the patient. This
may be skin surface or exposed bone or soft tissue. Though the data in a sin-
gle camera image is 2D, 3D information is potentially available from the use
of perspective. By taking multiple camera images from different positions, a
better estimation of 3D location can be achieved.
12.4.1.1
Perspective Calibration
The camera gives a perspective projection of 3D space into a 2D image. It is
standard to use the pinhole camera model (Figure 12.1). Calibration generally
involves taking a camera image of an object with high-contrast markings at
known 3D positions. The image is then transferred to a computer for analysis
using a framegrabber. The 2D positions of the markers are usually found
automatically. The corresponding 2D and 3D points are used as the input
data for the calibration.
In this section only the full perspective model will be discribed. Simpler
projection models have been proposed, such as weak or parallel projection,
which enable easier mathematical formulation at the expense of some inac-
curacy. The full perspective model is accurate in the presence of negligible
geometric distortion.
FIGURE 12.1
The general perspective projection model. The effective pinhole of the camera is at
O
. Pixel
coordinates in the image plane are (
U
,
V
) and (
u
,
v
) is the nearest point on the image plane
0
0
to
O
. A general 3D point (
x
,
y
,
z
) projects to (
u
,
v
) by Equation 12.1.
