tho-image mosaics,
/Radiometric Point
4rch. Phot. & Rem.
1.
hrer, and T. Fuchs,
tive to LiDAR for
Sens., Calgary, AB,
d K. Morin, 2011.
int Clouds Derived
'onference, Milwau-
sient Stereo Proces-
| Information. Proc.
v York.
by Semiglobal Mat-
sactions on Pattern
No. 2, pp. 328-341.
arge Format Digital
- Fernerkundung —
). 71-79.
nitial Results. Euro-
ww.ipf.tuwien.ac.at/
it LinzExperiences.
Zckardt, A., Hilbert,
os, E, Reulke, R,,
H Systems ADS40
Rem. Sens., Amster-
258-265.
chure, ZI Imaging.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
AN ACCURACY ASSESSMENT OF AUTOMATED PHOTOGRAMMETRIC
TECHNIQUES FOR 3D MODELING OF COMPLEX INTERIORS
Athanasios Georgantas!”, Mathieu Brédif!, Marc Pierrot-Desseilligny!?
g gny
!IGN, MATIS, 73 Avenue de Paris 94160, Saint-Mandé, France; Université Paris-Est
name.sumame @ign.fr
2Université Paris Descartes, CRIP5-SIP Lab, 45 rue de Saints Pères, 75006, Paris, France
Commission III/1
KEY WORDS: Close Range Photogrammetry, Bundle Adjustment, 3D Modelling, Automation, Accuracy, Open Source
ABSTRACT:
This paper presents a comparison of automatic photogrammetric techniques to terrestrial laser scanning for 3D modelling of complex
interior spaces. We try to evaluate the automated photogrammetric techniques not only in terms of their geometric quality compared to
laser scanning but also in terms of cost in money, acquisition and computational time. To this purpose we chose as test site a modern
building’s stairway. APERO/MICMAC ( ©IGN )which is an Open Source photogrammetric software was used for the production of
the 3D photogrammetric point cloud which was compared to the one acquired by a Leica Scanstation 2 laser scanner. After performing
various qualitative and quantitative controls we present the advantages and disadvantages of each 3D modelling method applied in a
complex interior of a modern building.
1 INTRODUCTION
Three dimensional models have become an essential tool for ex-
perts in various domains. They are used in urban and environ-
mental planning, in cultural heritage documentation, in build-
ing and infrastructure inspection, in industrial measurement and
reverse engineering applications, in film industry and in video
games and virtual reality applications. Various methods are used
in order to produce these models such as photogrammetry, laser
scanning and traditional surveying.
Terrestrial laser scanners (TLS) have been widely used since their
introduction in the production of such models. Their geometrical
accuracy and automation which was outperforming the classic
close range photogrammetric techniques made them very popu-
lar in various applications such as cultural heritage documenta-
tion and building inspection.
However multi-image matching which was introduced well be-
fore (Gruen and Baltsavias, 1998) the appearance of laser scan-
ners has become a very active research topic in both the pho-
togrammetric and computer vision communities. Even though
the two communities have been working almost independently
till the year 2000 (Forstner, 2009) this quickly changed as it be-
came clear that the combination of the techniques used by both
communities could lead to serious advances in the automation of
the close range photogrammetric procedures. The introduction
of tools like SIFT (Lowe, 1999) or MSER (Matas et al., 2002)
that can reliably extract dense features from overlapping images,
the automatic orientation of large blocks of images using only tie
points (Snavely et al., 2008) and the dense matching of oriented
images with the use of multi-correlation techniques and optimiza-
tion techniques (Furukawa and Ponce, 2010) have contributed a
lot to this direction. However it is not only the automation of
the algorithmic process that helped photogrammetry to become
once again attractive for 3D modelling. The advances in digital
cameras that led to the production of low cost high quality off
the shelf cameras was another important aspect that helped pho-
togrammetry to re-establish herself as a competent player in the
field of accurate 3D modelling.
Nowadays both TLS and photogrammetry can be used for the
23
production of 3D models even though they both present certain
advantages and disadvantages. On the one hand TLS offers high
automation on the procedure of acquiring dense 3D models of
high geometric accuracy but the costs of acquiring a laser scanner
and the software needed to manipulate the scanner and the cap-
tured data remains still very high. Another disadvantage of the
TLS is that the terrain acquisitions are usually time consuming
and the RGB information acquired is usually of low quality for di-
rect texturing. On the other hand even though detailed 3D capture
through photogrammetry is highly computational and ambigu-
ous, it presents certain advantages compared to laser scanning
such as its lower cost, its flexibility, its rapid acquisition times on
the field and the direct production of textured point clouds by us-
ing directly the RGB information from the acquisition’s images.
Nevertheless the geometric accuracy is strongly dependent by the
objects texture, the calibration of the camera, the use of control
points, the resolution of the images used and the network geom-
etry. So the question on which of the two methods offers the best
results in terms of different criteria still remains open.
In this paper we are presenting an accuracy assessment of 3D
point clouds of complex interiors produced with a fully auto-
mated open source photogrammetric software developed within
the IGN (French Mapping Agency). We are also interested in
defining the error sources in the different phases of a photogram-
metric acquisition and the reliability of a photogrammetric acqui-
sition in terms of field and office time compared to those with
a laser scanner. We have chosen as test site a building’s stair-
way. The stairway dataset was acquired with a Canon 5D Mark
II and a 8mm fish-eye lens. In order to be able to compare the
metric quality of our photogrammetric point cloud we have used
a Leica Scanstation2 laser scanner which provided us with a 3D
point cloud of known measurement uncertainty that were used as
a reference.
2 RELATED WORK
The introduction of laser scanners in the early 90s and their di-
rect competition with photogrammetry, analytical and digital, in
the field of 3D surface and object measurements has led to vari-
