Figure 4. Intersecting the line segment planes with the
XY-plane
— — t2
© CA ©
A 4 1
UA
4
Weighted sum of line segments
0 A
0 25 50 75 100 125 150 175
Intersection angle with XY-plane
Figure 5. Finding main directions (all images)
Figure 3. The segmentation result using Object Related Ima
building.
956
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
ge Segmentation (ORIS) on six image patches of one
3.5 Searching for the main structures
Having found the main directions in object space, i.e., the
vanishing directions in the images, a 3D clustering of the
line segments that contribute to these directions is carried
out. The clustering of intersecting line segments is done
for all images simultaneously. The aim of the evaluation
of the clusters is to determine the correspondence
between image lines (Fig. 6). Geometrical rules like
parallelity, perpendicularity and connectivity are
incorporated in the clustering algorithm.
In Fig. 7, all line segments belonging to a main
direction are projected on a cluster plane and in Fig. 8
only clusters that have at least five intersecting lines. The
cluster planes are evaluated using MDL to remove
contradictions and to find the most probable solution. The
found 3D lines are back-projected on an image in Fig. 9.