Graphics Reference
In-Depth Information
Fig. 5.18
Schematic description of the specular stereo algorithm
5.4.1 Iterative Scheme for Disparity Estimation
We utilise a correlation-based blockmatching stereo algorithm (cf. Sect.
1.5.2
) to ob-
tain depth information about the surface. The images are rectified to standard stereo
geometry, resulting in epipolar lines corresponding to the image rows (cf. Sect.
1.5
).
Directly applying the stereo algorithm to an image pair of a rough metallic surface
usually results in a fairly sparse disparity map due to limited texture, repeating pat-
terns, or a different appearance of corresponding surface parts in the stereo images
as a consequence of the strongly specular reflectance behaviour.
The coordinate systems of the two cameras are denoted by the indices
C
1
(left
camera) and
C
2
(right camera), the corresponding rectified coordinate systems by
R
1
(left rectified camera) and
R
2
(right rectified camera). The transformations be-
tween these coordinate systems and therefore also the viewing directions
C
1
v
1
and
C
1
v
2
of the cameras are known from the extrinsic camera calibration; e.g. let us
look at coordinate system
C
1
. We assume that the surface is illuminated with a point
light source situated at infinite distance in a direction given by the vector
C
1
s
.The
intensity and polarisation angle reflectance functions
R
I
and
R
Φ
areassumedtobe
known from a reference measurement. The proposed stereo image analysis method
for non-Lambertian surfaces is termed specular stereo. It consists of the following
steps (cf. also Fig.
5.18
).
1.
Compute a three-dimensional surface profile based on SfPRD
: A three-dimens-
ional surface profile is computed with the SfPRD method based on the intensity
and polarisation data of camera 1 and the depth points
C
1
r
(m
j
obtained by stereo
analysis, where
m
denotes the iteration cycle index. Each pixel is regarded as
Search WWH ::
Custom Search