Projected line
segment planes
2
di
M
X ™~ M *
ml Moa
True cluster 35 M E
SE AME M RA
A Pd. T
S t
+.) 6) False
A
; clusters
©
‘ 7€ P ad
—
s ^
o
x Accumulated
clusters
of one
Figure 7. A clustering with line segments from six
images, profile from house as overlay
ace, i.e., the
tering of the
ns is carried
ents is done
e evaluation
'espondence 4
| rules like ET
tivity are ur ne S.
! to a main : bendy
id in Fig. 8
g lines. The
to remove
lution. The à
in Fig. 9.
Figure 8. Clusters with at least five intersecting lines,
profile from house as overlay
957
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Figure 9. Found 3D lines projected on one image.
3.6 Finding 3D Planes
We are now facing a small number of reliable, horizontal
3D lines. The lines are in principle endless, but here
truncated to a reasonable object size. The lines may be
pair-wise parallel, or, when projected onto the XY-plane,
perpendicular. The pairs of lines that are parallel can
define a 3D plane. Each of these planes is a hypothesised
object plane, regardless if it overlaps other plane
hypothesis or not. Each of the plane hypothesis is back-
projected to each image, generating a 2D window. The
lines of the region boundaries that fall completely within
the window are analysed with respect to perpendicularity.
In practice, the window is enlarged somewhat (in the
example corresponding to appr. 0.5 m) to include lines
just nearby the window. Examples of one correct and one
false 3D plane hypothesis, projected onto one image and
including the lines enclosed by the windows are shown in
Fig. 10.
Figure 10. One correct plane window (left) and one false
(right) including the 2D lines from the image that are
circumvented by the respective window. The window is
somewhat enlarged to include nearby lines. The same
procedure is repeated for all plane hypothesis and all
images.