IMPROVING THE ACCURACY OF PHOTOGRAMMETRIC ABSOLUTE
ORIENTATION USING SURFACE MATCHING
S. J. Buckley **, J. P. Mills ^, H.L. Mitchell *
? School of Engineering, University of Newcastle, Newcastle, NSW 2308, Australia
simon.buckley, harvey.mitchell@newcastle.edu.au
b School of Civil Engineering and Geosciences, University of Newcastle upon Tyne, Newcastle, NEI 7RU, UK
j.p.millst@ncl.ac.uk
Commision II, WG I
KEY WORDS: Accuracy, Aerial, Digital, DEM/DTM, Matching, Orientation, Registration, Surface
ABSTRACT:
Research was conducted into improving the accuracy of photogrammetric absolute orientation, using a least squares surface
matching algorithm rather than conventional ground control points (GCPs). To ascertain the success of the developed algorithm, a
comparison between the two methods was carried out. Targets were laid at a test site, and near-vertical stereopairs of small format
imagery were captured using a Kodak DCS 660 camera and microlight platform, from three flying heights. Stereomodels for each
height were orientated using six GCPs as control, after which DEMs were extracted using automatic routines and compared with the
remaining targets.
For the surface matching orientation a. kinematic GPS DEM of the test area was collected, and image
stereomodels recreated — this time processed only to the relative orientation stage. DEMs were again extracted but were not yet in
the desired reference system, requiring development of the matching algorithm. As an alternative means of performing the
orientation, GCPs were bypassed and the GPS surface was used to register the unorientated DEMs. The matched surfaces were then
compared with the checkpoints, showing a higher correspondence to the checkpoints for all three flying heights. This suggests that
higher orientation precisions can be achieved with DEMS than using traditional GCPs, an especially important result for small format
photogrammetry where more images and more GCPs are required. Although this study focused on small format imagery of a single
area, implications on the wider discipline of photogrammetry are readily apparent.
1. INTRODUCTION
The conventional photogrammetric processing approach
required a minimum of three height and two plan ground
control points (GCPs) per stereopair to allow successful
absolute orientation to be carried out (Rosenholm and
Torlegard, 1988). Modern developments in digital
photogrammetric ^ workstations (DPWs) employ aerial
triangulation and simultaneous bundle adjustment to combine
the relative and absolute orientation stages into a single process,
requiring fewer ground control points, with automatically
measured tie points providing an efficient substitute to link
adjacent stereomodels. Despite this, for some terrain areas the
identification of hard features to be used as ground control may
be difficult, requiring logistically inefficient and expensive pre-
marking, or complex direct inertial systems as an alternative.
Examples of such terrain areas are coastlines, landslides,
glaciers and deserts, where the often dynamic nature of
occurring processes, and few natural or man-made hard features
make the collection of photocontrol the most inefficient and
least automated part of the processing chain (Schenk, 1999).
This stage also constitutes a significant monetary cost —
between 10% and 50% of a project’s expense (Warner et al,
1996; Wolf and Dewitt, 2000).
It is becoming increasingly common for digital elevation
models (DEMs - defined here as regular or irregularly
distributed point sets) to be the main end product of a survey
* Corresponding author.
project, from photogrammetry, airborne laser scanning (ALS),
terrestrial laser scanning and airborne Synthetic Aperture Radar
(SAR) Interferometry (InSAR). These surfaces may have use
in many disciplines and applications, for geomorphological
change detection, coastal erosion monitoring, flood prediction
and architecture to name but a few. Consequently, in these
projects it is the final DEM that is important, and this provides
an alternative processing strategy for photogrammetric DEMs.
Instead of using costly GCPs to provide registration to the
desired reference system, a photogrammetric DEM may be
orientated using an existing DEM and a surface matching
algorithm — in effect using a control surface to perform the
registration. This paper focuses on the accuracy of the ensuing
absolute orientation, by comparing the performance of
conventional orientations using GCPs and orientations using
surface matching, for DEMs produced from digital small format
photography captured from varying flying heights.
2. SURFACE MATCHING
Surface matching provides a common and fundamental problem
in computer vision (Zhang and Hebert, 1999); however, its use
in the spatial information discipline is intrinsically linked. The
general problem, given two free-form shapes or point sets, one
of which may be in a reference coordinate system and the other
of which may be in a model coordinate system, is to find the
International Arc
rigid transforma
correspondence |
finding the "opt
registers, the mo
distance between
apparent that a si
unorientated pho
reference DEM.
A popular choice
Closest Point (
designed to matc
primitives such a
information field
least. squares a
surfaces. Indeed,
methods, applied
most often in
implementation,
algorithms. For
method that mini
was carried out in
The surface ma
standard — seven-
commonly used i
the coordinates o;
(Wolf and Dew
procedure is com
be identifiable t
relating to the
collection techni
temporal change
acquisition of ea
used, the aim of t
that may then be
separations betwe
triangulated refer
minimised in the
transformation p
matching implem
disparate data, the
patch gradients tc
and Torlegard, 1¢
Mitchell and Cha
of these difficult
significant advan
potential for au
independent refer
unorientated DEM
The surface mat
stage in a coast
extracted from
photography (SF.
reference system.
determine the suc
Consequently, te;
accuracy achieva
approach using G
following sectio
processing and D
results and discus: