Graphics Reference
In-Depth Information
Fig. 1.18
100. The
solid con-
tours
have been extracted from the
right image
of the stereo pair, the
dashed contours
from the
left image
. Contour points on the
solid contour
for which a disparity value has been determined
are marked by
solid dots
. Only the contours obtained with binary threshold
θ
0
are shown. (
b
) Con-
tours extracted from absolute difference image. The
solid
(binary threshold
θ
0
(
a
) Contours extracted from absolute difference image with
θ
0
=
50) and the
dotted
(adaptive binary threshold derived from intensity histogram)
contours
have been extracted from the
right image
of the stereo pair, the
dashed and the dash-dotted contours
in a corresponding manner
from the
left image
.The
solid and the dashed contours
are partially determined by shadow struc-
tures on the floor rather than by 'real' object contours. Contour points for which a disparity value
has been determined are marked by
solid dots
(U
(j)
2
=
,V
(j)
2
1
,...,n
2
. The extracted contours are smoothed by B-spline in-
terpolation. Subsequently, we calculate the
u
(column) coordinates at which the
interpolated contour intersects the rows in the
v
range covered by the corresponding
B-spline in order to obtain subpixel accuracy (cf. Fig.
1.18
). Hence, the contours
are represented as sets
), j
=
c
(m)
2
,a
c
(n)
{
}
and
{
1
,b
}
of points for the left and the right image,
respectively:
1
,b
=
¯
1
,b
c
(n)
u
(n)
v
(n)
1
,b
,
¯
(1.105)
=
¯
2
,a
,
c
(m)
2
,a
u
(m)
v
(m)
1
,a
,
¯
where
a
=
1
,...,n
2
are the blob indices in the images and
m
and
n
the point indices in the respective contours. The values
1
,...,n
1
and
b
=
u
(n)
1
,b
u
(m)
2
,a
¯
and
¯
are real
v
(n)
1
,b
v
(m)
2
,a
numbers, while the values
and
denote image rows and thus are integer
numbers.
In the following, a pair of blobs mutually assigned by the previously described
procedure is regarded. For each epipolar line in the range covered by both blobs, the
numbers of intersections between the contour and the epipolar line are calculated for
both images. These intersection counts are denoted by
e
1
(v)
for the right and
e
2
(v)
for the left image of the stereo pair. Along with these values, the
u
coordinates of the
epipolar intersections,
u
(i)
1
1
,...,e
1
(v)
for the right image and
u
(j)
=
1
,...,e
2
(v)
for the left image, are determined. They are known to subpixel accuracy
(v)
,
i
=
(v)
,
j
2
Search WWH ::
Custom Search