International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Prior to the aerial survey being carried out, fifty-four one-metre
diameter white targets were laid, distributed to ensure that an
adequate number would be visible within the image overlap.
These were coordinated using a Leica TCR307 total station,
providing data with 7” angle measurement precision and 2 mm
+2 ppm. Two control markers were monumented and observed
in a static Global Positioning System (GPS) network, allowing
the various acquisition techniques to be registered to the UTM
coordinate system, a plane projection of WGS-84. These
control points formed the total station set-ups for observing the
ground targets. Convergent imagery of the targets was captured
during the aerial sortie, and used to calibrate the DCS 660
camera in a self-calibrating bundle adjustment, resulting in a
root mean square (RMS) error of 0.35 uum and a relative
network precision of 1:42,000.
3.3 Conventional Orientation
Imagery was processed in Leica Geosystems' SOCET SET
version 4.3.1 (now owned wholly by BAE Systems), a digital
photogrammetric workstation allowing image orientation and
DEM extraction. A model was conventionally orientated in
SOCET SET using a stereopair of images captured from 600 m
and six of the targets as GCPs; these were distributed
throughout the stereocoverage to attain a strong adjustment
solution — less than 0.01 m coordinate residuals and an image
residual of 0.38 pixels. At 1:22,000 and with a 60% overlap,
this configuration provided a B/H ratio of 0.4 resulting in an
expected heighting precision of 0.35 m (Light, 2001).
Following orientation, a DEM was created using the automatic
terrain extraction (ATE) facility of the DPW. Least squarcs
correlation is used for this task, and the adaptive parameter
strategy was chosen for extracting the DEMs. This is a more
sophisticated algorithm than the non-adaptive methods,
parameter strategies of which are determined according to user-
specified terrain types. Instead of requiring a single parameter
strategy for each DEM, regardless of terrain type, the adaptive
ATE performs epipolar resampling and Y-parallax removal, as
well as examining image and terrain content to modify
matching parameters on-the-fly. Using this strategy, a
triangular irregular network (TIN) DEM of the test area was
created, with progressive sampling based on a 2 m grid.
Identical methodology was carried out for the imagery captured
from 270 m and 450 m, the only difference being that, because
the lower flying reduced the ground coverage of the images,
fewer check targets were visible in an image stereopair —
detrimental for the later accuracy comparisons. This was
resolved by using a strip of three images for the 450 m model,
containing 24 targets, and four images for the 270 m model,
containing 14 targets (excluding those targets used as GCPs).
Following photogrammetric processing, a DEM of the cliff test
area existed for each of the three flying heights where imagery
was acquired. These conventionally controlled DEMs were
then compared with the remaining ground check targets to give
an estimate of the precision of the photogrammetric surfaces.
Measures defining DEM accuracy arc essential, but have not
yet reached a point where a standard method exists to determine
the correspondence between a digital terrain representation and
the ‘true’ surface. Consequently, there are few guidelines for
practitioners (Flotron and Koelbl, 2000). A common means to
define accuracy has traditionally been descriptive statistics
based on the difference in height between the DEM surface and
n check measurements. Because the check data is usually point
or surface based, it is unlikely that conjugate points will be
available, and the DEM height value is therefore interpolated at
the planimetric positions of the validating data. The mean,
standard deviation and range of the extreme Z differences can
then be determined (e.g. Shearer, 1990). Z differences were
interpolated between the photogrammetric DEMs and the
checkpoints contained within the extents of each surface.
Standard deviations of these were calculated, with results
presented in Table 1 and Figure 2.
Flying height 270 m 450 m 600 m
Expected c (m) 0.153 0.255 0.350
Conventional o(m) 0.401 0.438 0.496
Table 1. Expected and conventionally controlled heighting
precisions for various flying heights
0.6
|
0.5 Voi
E ere RR -
E 04 B————
: |
5 9
i
i
à 92
0.1
i
270 450 de
Flvina heiaht (m)
| —@— expected ii Convertiorial
Figure 2. Effect of flying height on DEM precision
Examination of these reveals that worse than expected
precisions resulted for all three flying heights used, implying
that such a formula is too optimistic for the format or
configuration of the photography. However, for the three
heights used, the expected precision increases linearly with
height whereas the observed precision line has a lower gradient,
suggesting that the formulae may represent higher altitude data
better.
4. SURFACE MATCHING ORIENTATION
The second stage of this research involved the use of surface
matching to perform the critical registration of the small format
photogrammetric DEMs to the desired coordinate system,
instead of using ground control points. For this stage a second,
independent surface was required, and this was achieved using
kinematic GPS.
4.1 Kinematic GPS
Since its inauguration in 1994, GPS has been used primarily in
the observation and monitoring of static point networks.
However, with advances in technology and processing
techniques, it is now possible to reliably achieve a high
measurement precision without the long occupation of single
points.
processing individual data observations relative to a base station
sited over a known point. Because of the highly accurate point
data, often quoted to around the 0.010 m level (e.g. Hofmann-
Wellenhof et al., 2001), kinematic GPS has become popular for
recording the trajectory of a moving receiver. A critical
Kinematic GPS has become an efficient means of
International Arc
component of pi
between dual fr
reference and r«
conducted in this
by the introducti
ambiguities “on-t!
A wireframe DE
using kinematic
receiver as it tre
(Figure. 3). -To
minimise change
standard detail po
used (Buckley ant
kinematic OTF p
relatively coarse,
coastline test s
Repeatability test
a baseline at the |
a value of 0.014
height precision t
individual points -
undulations, veget
Figure 3. Perspe
4.2 Registration
Digital SFAP of
DPW, for each o
followed a similar
one exception: noi
SOCET SET use
relative and abs
presented a prol
controlling photog
using surface mat
adjustment require
adjustment to ot
Consequently, thr
Stereopair (and firs
and 450m flying
mapping. | Obvia
detrimental to the
tie points were 1
employed to increa
Once the simultane
DEMs of the same
were extracted usi