Bitelli, Gabrielle
introduce unacceptable errors due to radial image point displacements (up to theoretical values of Ap=10.9 mm on the
image): differential rectification is required, which leads to creation of the orthophoto. In such cases, it is essential tg
have 3D mesh models of the facade, on which carry out differential projection of the individual elements of the image,
The 3D model of the object may be made manually, in semi-automatic or fully automatic mode. The following section
describes a set of tests for automatic DSM generation.
3 AUTOMATIC DSM GENERATION TESTS
The aim of matching in photogrammetry is to define the three-dimensional object coordinates from images. The points
in the object-space are related to the points in the image system by means of the collinearity condition.
However, to solve image matching, that is to say, to establish correspondence between images or between images and a
model, there is not one solution, but a variety of strategies which are often combined (Hahn, Fórstner, 1988; Heipke,
1996; Helava, 1988; Kraus, 1997). Yet the matching problem cannot always be solved and even the most sophisticated
algorithms recently produced may fail or supply ambiguous solutions. For example, for a given point on an image the
corresponding homologous point on other image(s) may not exist or may be in an ambiguous position due to repeated
geometric patterns or poor image information content (texture).
A variety of matching algorithms presented in literature is implemented in digital photogrammetry stations. The
research in question made use of the low cost StereoView software, running on PC platform, made by Menci Software
stl (http://www.menci.com) and distributed by Nikon Instruments spa. The software uses for the image matching a
mixed area/feature based algorithm with the matching method known as VLL Matching (Vertical Line Locus; Kraus,
1993). Thus, the program can create a 3D model of the object studied, starting with a pair of digital stereoscopic images
and known object coordinates. However, automatic point positioning does require a minimum user intervention. The
images are matched on a regular grid of the model, defined in advance by the user. First, at least 3 three-dimensional
points must be positioned manually, to allow definition of a reference surface for the start of matching. The automatic
matching process starts from the starting point, that is to say, from the three-dimensional point located in the working
area which has the greatest correlation coefficient. It may be a control point or a three-dimensional point entered by the
operator (therefore, with maximum correlation coefficient). The final product of matching is a DSM, which is easily
manipulated to create a 3D raster model of the object, of which the orthophoto constitutes a special view (orthogonal).
To create a good quality DSM, suitably defining the automatic matching parameters and minimising editing before and
after, the photograms are first scanned using a DTP (Desk Top Publishing) scanner with 1200 dpi resolution,
guaranteeing a object pixel size of approx. 4 mm (table 1).
The scanned images (figure 2) are corrected to remove the geometric deformation errors introduced during the
acquisition stage, using a special calibration module (SVScan), supplied in addition to the StereoView software.
After the internal orientation, the photograms are externally oriented by means of bundle adjustment, simultaneously,
on the basis of the control and observation points. The orientation residuals in the various models show an average
standard deviation value in the three coordinates of 2.8 mm.
Camera P31 Wild Image resolution 1200 dpi
Lens 99.39 mm Image scale 1:190
Film size 90 x 120 mm | Pixel image (25400:1200)=21.67u
n. images / n. stereomodels 4/2 Pixel on the ground (21.67*190)= 4mm
Table 1 — Main characteristics of primary data acquisition and scanning.
Figure 2 — Photograms of the Zabbar Gate, test object.
64 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.
The
phot
7 DS
each
stere(
refer
origi
Com]
DSM
pixel:
basic
pixel:
the ol
size i
Ther
analy
Table
seven
percer
differ:
The p
digital
- thet
lower
- the 1
search
centra
minim
- thes
conclu
statisti
since t
empha
Theref
With le
The te
fundar
Furthe
that th
Zones '
The de
increas
distrib: