ignals,
achine
. Scale
d Ma-
, Pho-
ol. 58,
rds an
pairs,
3, Vol.
| Con-
edings
ntelli-
'orems
ntelli-
chnol-
Feature Matching for Outer Orientation of Single
Images Using 3-D Wireframe Controlpoints
Wolfgang Schickler
Institute of Photogrammetry
University of Bonn, FRG
Abstract
The paper presents an automatic model-based outer orientation (AMOR) of a single image using
pattern recognition techniques. Models are 3-D wireframe models of buildings. The development of
AMOR is part of an effort of the Survey Department Bonn, to automate the production of digital
orthophoto maps. Extensive tests confirm the high reliability of the approach, and also the internal
decisive measures for selfdiagnosis, and suggest the matching technique to be widely applicable.
Key words : Model-based object location, automatic outer orientation, 3-D matching,
maximum likelihood estimators, selfdiagnosis.
1 Motivation
Digital orthophoto will soon replace their analog ancestor
due to their versatility in use, their higher quality and cer-
tainly also lower costs. They will not only serve as a flexible
basis for planning purposes, but also support quick map up-
date.
Automatic orthophoto production needs several components
to be available. À scanner and a plotter for digital input and
output, a digital elevation model (DEM), information about
the outer orientation and a powerful workstation. While con-
cepts for differential rectification of digital images, programs
for deriving DEM or possibly the DEM itself are available
and integrated into systems, automatic procedures for de-
terming the outer orientation are rare [e.g. MALMSTROM H.
86 |
In case of a periodic update of orthophoto maps it is no lon-
ger necessary to perform an aerial triangulation each time, if
there are automatically identifyable control points are availa-
ble. Then the determination of the 6 orientation parameters
may be achieved by individual spacial resections, which is
unless costly. This was the motivation of the Survey Depart-
ment Bonn to build up a control point database, consisting
of more than 20 000 natural contol points, mostly houses al-
ready more than 10 years ago. Within their up to now analog
orthophoto produktion these control points were measured
manually. In 1993 they will start to produce digital ortho-
photos. This change was the reason to develop a new con-
cept concerning the automation of the whole produktion.
The computer having access to the digital image data allows
to replace the manuall identification and measuring process
by an automatic procedure allowing the setup of a produc-
tion line fully independent of human observer. Whereas the
rectification of the digital images causes no problem with re-
spect full automation the identification of externally given
objects, i. e. control points requires nothing more than a re-
liable image understanding procedure.
591
As pure correlation, thus low level matching, procedures are
not feasible in this case, due to the threedimensionality of
the control points causing strong distortions, possibly oc-
clusions, only an orientation procedure based on features is
able to solve the matching task. More over, as the produk-
tion has to be autonomous, the orientation procedure needs
to contain a feature for selfdiagnosis, reliably predicting the
success or failure of the procedure, as in (the unlikely) case
of a failed orientation no rectification needs to be performed.
This paper describes the main features of the program
AMOR for the automatic model-based orientation of aerial
images. Though the actual procedure is optimized with re-
spect to this task, the concept and most of the modules may
be transferred to other applications.
2 Overview
The orthophoto production of the Survey Department Nord
Rhein Westfalen is based on images of approximate scale 1 :
12 000 which are digitized with a pixel size of 25um, a DEM
and on a control point database. The control points, mostly
roofs of houses, are given by orthographic sketches and 3D-
coordinates of 2 distinct points. The sketches are used for
the manual localisation of the control point buildings in the
image. The density of the control points is high enough that
5 to 8 control points are available for outer orientation.
The possibility of using these 2-D sketches as models for
object location based on uncertain models [FÖRSTNER W.
88 ; SESTER M. / FÖRSTNER W. 89] has been proved. The
interpretation of the 2-D sketches lead to 3-D discriptions of
the house roofs [SCHICKLER W. 89]. As the sketches have
shown to be not fully in scale, the location process has been
set up and realized in a way which took the uncertainty of the
models into account, treating them as stochastical variables.
Against previous assumptions, the sketches have shown to
be not suitable for this purpose.
