Graphics Reference
In-Depth Information
Fig. 1.24
Mismatches in the
marked area
produced by different stereo methods.
Light grey points
:
correct disparities;
dark grey points
: incorrect disparities. The three-dimensional points are shown
in disparity space, where
u
and
v
are the image coordinates of a pixel in the
right image
and
d
denotes the associated disparity value. (
a
)
Right image
of a stereo pair, area with repetitive
structures marked in
light grey
.(
b
) Feature-based method (spacetime stereo algorithm; Schmidt et
al.,
2007
). (
c
) Intensity-based method (blockmatching stereo algorithm; Horn,
1986
). (
d
) Global
method (semi-global matching algorithm; Hirschmüller,
2005
)
model pose is used to perform a refined correspondence analysis. This analysis is
governed by a cost function which takes into account the distances of the three-
dimensional points from the model.
In this section, the stereo approach introduced by Schmidt et al. (
2007
) based on
spatio-temporal local intensity modelling (cf. Sect.
1.5.2.5
) is used, where different
stereo constraints can be taken into account to find the correct matches. A com-
parison of the results of the well-known uniqueness (Marr and Poggio,
1979
) and
ordering constraints (Baker and Binford,
1981
) (cf. Sect.
1.5.2
) is performed. Fur-
thermore, a model-based stereo method similar to the approach by Tonko and Nagel
(
2000
) is regarded for comparison. It uses the model surface and the pose parame-
ters to obtain the homographies that determine corresponding image points in both
images. Image windows of predefined size around a given number of points in one
stereo image are reprojected into the other image based on these homographies in-
duced by the model surface. The similarities between corresponding image windows
are maximised by variation of the pose parameters.
Search WWH ::
Custom Search