Information Technology Reference
In-Depth Information
(a) LC matching with intensity map (b) LC matching with
|R|
-based map
Fig. 5.
Result of Local Contour matching with different edge maps
3.3 New Edge Map
As it was also written in Section 2.2 intensity based edge maps are sensitive to
illumination changes. This problem also occurred when comparing local features:
changeless places were declared as newly built objects. Therefore we used the
Harris based feature map, denoted by
R
old
and
R
new
in Section 2.2:
f
|R|
(
x, y
)=
G
σ
(
x, y
)
∗|
R
(
x, y
)
|
(5)
|
R
|
Detected contours are smoother and more robust in case of the
function. We
benefit from this smoothness, as contours can be distinguished easier. However, as
there might be no real contour in the neighbourhood of the keypoints, AC-method
is only used for exploiting the local information to get low-dimensional descriptor,
therefore significance of accuracy is overshadowed by eciency of comparison. The
detected points based on the
|
R
|
-function can be seen in Figure 5(b).
3.4 Enhancing the Number of Saliency Points
After selecting the saliency points (or keypoints) indicating change, we now have
to enhance the number of keypoints. Therefore we are looking for saliency points
that are not presented in the older image, but exists on the newer one. We call for
the Harris corner detection method again. By calculating Harris corner points
for older and newer image as well, an arbitrary
q
i
=(
x
i
,y
i
) point is selected if
it satisfies all of the following conditions:
(1.)
q
i
∈
H
new
(2.)
q
i
/
H
old
(3.)
d
(
q
i
,p
j
)
<
4
H
new
and
H
old
are the sets of Harris keypoints generated in the newer and older
image,
d
(
q
i
,p
j
) is the Euclidean distance of
q
i
and
p
j
,where
p
j
denotes the point
with smallest Euclidean-distance to
q
i
selected from
H
old
.
New points are searched iteratively, with
4
= 10 condition. Here,
4
depends
on the resolution of the image and on the size of buildings. If resolution is smaller,
than
4
has to be chosen as a smaller value.
Figure 6 shows the enhanced number of keypoints.
∈
Search WWH ::
Custom Search