OLUTE
alia
e, NEI 7RU, UK
a least squares surface
developed algorithm, a
reopairs of small format
Stereomodels for each
; and compared with the
s collected, and image
cted but were not yet in
cans of performing the
iched surfaces were then
ghts. This suggests that
1t result for small format
rmat imagery of a single
e laser scanning (ALS),
ynthetic Aperture Radar
| surfaces may have use
, for geomorphological
itoring, flood prediction
Consequently, in these
ortant, and this provides
iotogrammetric DEMs.
vide registration to the
mmetric DEM may be
nd a surface matching
surface to perform the
accuracy of the ensuing
y the performance of
; and orientations using
from digital small format
ing heights.
CHING
ind fundamental problem
, 1999); however, its use
intrinsically linked. The
shapes or point sets, one
rate system and the other
ite system, is to find the
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
rigid transformation relating the two surfaces, to establish
correspondence (Besl and McKay, 1992). This relates to
finding the “optimal rotation and translation that aligns, or
registers, the model shape and the data shape minimising the
distance between the shapes" (Besl and McKay, 1992). It is
apparent that a similar problem exists in this research, where an
unorientated photogrammetric DEM is to be registered using a
reference DEM.
A popular choice in the computer vision field is the Iterative
Closest Point (ICP) algorithm (Besl and McKay, 1992),
designed to match not only surfaces but also other geometric
primitives such as line segments and curves. Within the spatial
information field, matching algorithms have tended towards
least squares adjustments, minimising quantities between
surfaces. Indeed, Mitchell and Chadwick (1999) argue that such
methods, applied to the relatively simple 24D DEMs found
most often in surveying, provide a more suitable
implementation, without loss of accuracy, than the ICP-style
algorithms. For this reason, development of a least squares
method that minimises the vertical differences between DEMs
was carried out in this research.
The surface matching approach adopted is based on the
standard seven-parameter 3D conformal transformation,
commonly used in photogrammetry and surveying, that relates
the coordinates of control points in different coordinate systems
(Wolf and Dewitt, 2000). With the use of surfaces, the
procedure is complicated by the fact that no control points may
be identifiable to carry out the transformation, for reasons
relating to the distinct point distributions; quantities; data
collection techniques used, with associated accuracies; and
temporal changes that may have occurred between the
acquisition of each dataset. Instead of control points being
used, the aim of the method is to find conjugate surface patches
that may then be used to carry out the transformation. Vertical
separations between the points of the unorientated surface and a
triangulated reference DEM are therefore computed, which are
minimised in the iterative least squares procedure, resulting in
transformation parameter estimates. Complications to the
matching implementation, relating to the use of irregular and
disparate data, the non-linearity of the solution, and the need for
patch gradients to exist in multiple directions (e.g. Rosenholm
and Torlegärd, 1988), are evident and are discussed further in
Mitchell and Chadwick (1999) and Mills et al. (2003). In spite
of these difficulties, this surface matching algorithm offers
significant advantages from the high level of redundancy, the
potential for automation and, importantly, by having an
independent reference surface that allows the accuracy of the
unorientated DEM to be validated.
3. EXPERIMENTATION
The surface matching method formed the critical orientation
stage in a coastal zone monitoring study, allowing DEMs
extracted from a strip of digital small format digital
photography (SFAP) to be effectively registered to a global
reference system. As part of this study, it was necessary to
determine the success of the algorithm and its implementation.
Consequently, testing was conducted to compare the DEM
accuracy achievable using both the conventional orientation
approach using GCPs, and the surface matching technique. The
following sections therefore detail the data collection,
processing and DEM extraction for both methods, as well as
results and discussion.
3.1 Digital Small Format Aerial Photogrammetry
Digital SFAP was chosen as the primary photogrammetric
acquisition technique because of its cost effectiveness and
speed of processing, especially for the single image strip
required for coastline coverage. To further speed up data
collection, the digital camera was mounted on a microlight
platform, allowing rapid scrambling and a larger weather
window than possible with a standard survey aircraft (Warner et
al., 1996). A significant limitation associated with SFAP is the
smaller ground coverage in each image, caused by the film size
or dimensions of the charge-coupled device (CCD) in a digital
camera. Combined with a focal length far shorter than that of
standard large format cameras makes for an exorbitant increase
in the amount of images needed to provide stereocoverage
(Warner et al, 1996). With the increase in images comes the
requirement for an increase in GCPs to provide an accurate
absolute orientation, making SFAP seem impractical for
anything other than the smallest areas. Hence the value of the
surface matching as an alternative orientation technique is
demonstrated.
3.2 Test Area and Data Collection
The area chosen for this study was the coastline of Filey Bay,
North Yorkshire, UK, a sensitive environmental area with
ongoing coastal erosion. For this experiment, a small section of
the bay was chosen, comprising a flat beach, gently sloping cliff
(rising to around 40 m), and grassed cliff top car park (Figure
1).
Figure 1. Orthophoto of Filey Bay test site, taken using
DCS 660. Area is approximately 200 x 200 m
Near-vertical stereo aerial photography of this test site was
acquired on 10 August, 2001, using a Kodak DCS 660 single
lens reflex (SLR) digital camera. This camera is one in a line of
high-resolution — (6 megapixel) cameras already used
successfully by the photogrammetric community (e.g. Maas and
Kersten, 1997; Chandler et aL, 2001). The camera was
mounted on a Thruster T600 Sprint microlight platform, the
lens fixed on the infinity setting and the aperture priority mode
set, ensuring an average shutter speed of 1/800 s at ISO200. To
investigate the heighting precision of this photogrammetric
configuration, imagery of the test area was captured from
varying flying heights: 270 m (900 ft; 1:9600 scale), 450 m
(1500 ft; 1:16,000 scale) and 600 m (2000 ft; 1:22,000 scale).
AT BEER