CITA 2003 XIX’ 1 ' International Symposium, 30 September - 04 October, 2003, Antalya, Turkey
668
of two ‘arches’ are presented in Fig. 6, along with the original
images. Perspective deformations have been removed, thanks to
the adopted modeling scheme. Finally, Fig. 7 gives the full 3D
model of the area and the resulting orthomosaic.
Figure 6. Images (left) and corresponding orthoimages (right).
Figure 7. Shaded 3D photogrammetric model and orthomosaic.
These examples, also founded on the experiences from previous
projects presented in Mavromati et al. (2002), indicate that the
adopted strategy is suitable for geometric image transformations
of both geometric accuracy and visual quality. As pointed out
above, object location and surroundings actually ruled out any
thought of experimenting here with laser scanning. To this end,
a site was surveyed which had already been fully mapped pho-
togrammetrically (Karras et al., 1999; Mavromati et al., 2002).
4. EXPERIMENTS WITH LASER SCANNING
For this archaic site in Athens, 7 images were chosen which had
been acquired vertically with a small format camera and 28 mm
lens (scale 1:1100). The mean RMS error in XYZ for 100 control
points was 3.7 cm (five times smaller than that of the solution
without correction of lens distortion). The site has very irregular
relief w.ith successive vertical ‘falls’ and a marked slope. Photo
grammetric point and breakline collection was performed in the
mode described above. The resulting surface model and the or
thomosaic were very satisfactory indeed. In Fig. 8 one can see a
shaded view of the 3D model.
W
Figure 8. Part of the shaded 3D photogrammetric model.
Fig. 9 shows an image of the object together with a further view
of the shaded model showing the success of reconstruction.
Figure 9. View of the object and shaded 3D model.
The site was surveyed using a CYRAX 2500 scanner at a resolu
tion of about 5 mm. A total of four stations were used, of course
not enough for wholly capturing the object but sufficient for the
experiment. Point clouds were automatically triangulated within
a commercial software package, to provide a direct (‘unedited’)
support for orthoprojection, which would then be compared to
the product from the photogrammetric surface model. While all
relatively smoothly shaped areas were orthoimaged in practical
ly identical manner from both sources, this was not the case at
the edges. Indeed, ‘vertical’ triangles were slightly deformed, as
is clearly seen in Fig. 10. This resulted in image ‘blurring’ and
‘erosion’ in the vicinity of discontinuities. Examples are given
in Fig. 11. As noted by Bitelli et al. (2002), use of laser-derived
models for orthophoto generation does not directly give satisfy
ing results (depending on shape, data density etc.); thus editing
of the model is needed, particularly as regards discontinuities.
