ence
verse
The
ig for
im on
ristol,
ctural
ping
5 and
18.
edge
om of
, Vol.
metric
yundle
3, Part
cts for
ol. 33,
xplicit
In: 6^
ıtterns.
space
grating
'achine
ictively
gs of
Vienna,
. single
nal of
4-368.
Bundle Orientation and 3-D Object Reconstruction from Multiple-Station Panoramic
Imagery
Thomas Luhmann, Werner Tecklenburg
University of Applied Sciences, Institute for Applied Photogrammetry and Geoinformatics,
Ofener Str. 16, D-26121 Oldenburg, Germany - (luhmann, tecklenburg)@th-oldenburg.de
Commission V, WG V/2
KEY WORDS: Close Range Photogrammetry, Bundle adjustment, Orientation, Panorama image, 3-D reconstruction
ABSTRACT:
The paper presents a new method for the precise mosaicking and orientation of multiple images from two or more camera stations
into cylindrical panoramic views. The objective is to generate panoramic images with high-precision orientation data in order to
extract 3-D object information. The first part of this paper reports on a method for distortion-free stitching of standard frame imagery
that generates panorama views of higher quality than it is qchieved by normal off-the-shelf panorama stitching programs. The second
part addresses the problem of photogrammetric orientation of multiple station panoramas that is solved by a modifed bundle
adjustment approach. Finally 3-D object information can be extracted from oriented panoramas in order to provide a cost-effective
way of 3-D modeling of the interior of rooms and buildings. Hence, facility management and architectural applications can be
identified as prior application areas for the presented approach.
1. INTRODUCTION
In conjunction with the use of modern information technology
the three-dimensional surveying and reconstruction of complex
industrial facilities and buldings are of growing importance.
Digital 3-D models of existing objects are already used for
planning pruposes, e.g. in architecture. However, the bandwidth
of practical applications is much wider. As an example, in plant
design precise geometric models are requested for the
documentation of pipe structures, or the replacement of
machine parts or facility components. The renovation or
redesign of complex buildings such as railway stations,
churches etc. is another example.
In the growing area of facility management three-dimensional
geometry data is also requested with increasing demand. Here,
building information systems are used where spatial object
information plays a major role.
Digital panoramic imagery (360?-views) has been used
increasingly over the last few years. However, their use is
mainly restricted to viewing impressions. A number of
commercial panorama viewers are available as plug-ins for
internet browsers in order to provide high quaility all-around
views of local environments, e.g. for touristic applications.
Usually these panoramas can not be used for 3-D object
reconstruction due to the lack of geometric camera models.
The here presented approach for the generation and orientation
of geometrically exact panoramas serves as a basis for the
measurement of 3-D information of the interior of rooms. The
defined image geometry (e.g. cylindrical projection) allows for
3-D object reconstruction in analogy to standard methods of
close-range photogrammetry.
2. STANDARD METHODS OF PANORAMA IMAGE
GENERATION
Basically two different approaches for the generation of
panoramic views can be distinguished:
2.1 Image formation with rotating line-scanning cameras
Rotating line-scanning cameras acquire a panorama by a
complete 360? rotation from one camera station. As examples,
the panorama cameras KARLINE (Dr. Clauf, Zwónitz) and
EYESCAN (KST, Dresden) can be mentioned (Fig. 1).
Fig. 1: Rotating line-scanning cameras EYESCAN and
KARLINE
During image acquisition both cameras are rotating in equal
angular steps around their vertical axis while the integrated
CCD line sensor is continously registering (Fig.2). A
geometrically exact panorama is created if the imaging
characteristics of the lens are known. Different approaches for
the photogrammetric evaluation of rotating line-scanner
cameras are discussed by Lisowski & Wiedemann (1999) and
Scheele et al (2001).
—181—
