. Part B3. Istanbul 2004
onjugate points will be
therefore interpolated at
iting data. The mean,
treme Z differences can
0). Z differences were
netric DEMs and the
tents of each surface.
alculated, with results
450 m 600 m
0.255 0.350
0.438 0.496
y controlled heighting
ng heights
600
L (m)
--- conventional
on DEM precision
worse than expected
; heights used, implying
stic for the format or
However, for the three
increases linearly with
line has a lower gradient,
esent higher altitude data
ORIENTATION
olved the use of surface
'ation of the small format
ired coordinate system,
For this stage a second,
| this was achieved using
as been used primarily in
static point networks.
nology and processing
reliably achieve a high
ong occupation of single
e an efficient means of
s relative to a base station
“the highly accurate point
0 m level (e.g. Hofmann-
S has become popular for
ng receiver. A critical
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
component of processing involves resolving the ambiguities
between dual frequency carrier phase measurements of the
reference and roving receivers. Much research has been
conducted in this area, and now the process has been simplified
by the introduction of sophisticated algorithms that determine
ambiguities “on-the-fly” (OTF).
A wireframe DEM of the Filey Bay test area was collected
using kinematic GPS, by tracking the position of a roving
receiver as it traversed breaks in slope and terrain profiles
(Figure 3). To facilitate data collection, and to attempt to
minimise changes in the antenna heights, the GPSycle — a
standard detail pole with a mountain bike wheel attached — was
used (Buckley and Mills, 2000). This data was processed using
kinematic OTF phase processing. The resulting DEM was a
relatively coarse, but highly accurate, representation of the
coastline test site, comprising strings of data points.
Repeatability testing of the kinematic GPS configuration, using
a baseline at the University of Newcastle upon Tyne, indicated
a value of 0.014 m was a more realistic indicator of DEM
height precision than the 0.010 m reported in the literature for
individual points — taking no account of factors such as terrain
undulations, vegetation or point distribution.
Figure 3. Perspective view of GPS DEM (200 x 200 m area)
4.2 Registration Using Matching
Digital SFAP of the Filey test area was reprocessed in the
DPW, for each of the three flying heights. The procedure
followed a similar route as for the conventional data but with
one exception: none of the ground targets were used as GCPs.
SOCET SET uses a single bundle adjustment to perform
relative and absolute orientation simultaneously. This
presented a problem for the proposed methodology of
controlling photogrammetric DEMs to the reference system
using surface matching rather than GCPs, as the bundle block
adjustment required a minimum of three control points in the
adjustment to obtain a solution for relative orientation.
Consequently, three ‘pseudo’ GCPs were measured in the
stereopair (and first stereopair of the image strip for the 270 m
and 450m flying heights), roughly scaled from existing
mapping. Obviously, the accuracy of the mapping was
detrimental to the quality of this control. A number of manual
tie points were measured before automatic matching was
employed to increase redundancy.
Once the simultaneous bundle adjustment had been carried out,
DEMS of the same area as used for the conventional orientation
were extracted using the ATE. Following DEM processing,
surface models were created, but were not yet registered to the
UTM coordinate system of the GPS surface. Therefore
matching was required to recover the transformation parameters
necessary to register the two models.
Pre-match processing of the GPS DEM was carried out to
ensure that the extents of the surface roughly corresponded
with, and were slightly larger than, the photogrammetric DEMs.
Only points where the ambiguity resolution was fixed were
incorporated into the DEM, creating the best possible surface.
A feature of the kinematic OTF phase processing is the high
data capture rate needed for successful ambiguity resolution;
however, this creates problems with the distribution of points in
the final DEM, with many points in the profile direction but few
between profiles, resulting in long, thin triangles and mass point
clusters in the TIN surface. The data were thinned to reduce the
observation rate and, additionally, the distance between points
was examined to ensure the existence of mass points, such as
where the roving GPS receiver was stationary for multiple
epochs, was eliminated. This helped give a more even
triangulation with more equilateral triangles — useful during the
search for conjugate surface patches in the least squares
matching algorithm.
Because the imagery from the different flying heights were
controlled using the 'pseudo' control points, it was expected
that initial approximations of one for scale and zero for each of
the rotations and translations would be suitable as initial
parameter approximations for the least squares solution.
Indeed, this was true of the 600 m and 450 m DEMSs, with only
reasonably small parameters found - suggesting that the
‘pseudo’ control points used were close to their true coordinate
values (Table 2). However, the large parameter corrections
seen in the 270 m match show that for this model a poor initial
absolute orientation was calculated in the DPW, resolved using
surface matching.
Match result 600 m 450 m 270 m
Outlier tol. 1m 1m 1m
Translation 1.407 £0.315 0.713 0.267 21.170 £0.557
(Xy 7m -0.913 0.233 -4.206 +0.157 -1.896 +0.296
0.863 0.107 -0.756 0.100 4.035 +0 175
Scale 0.996 +0.001 1.000 0.001 0.939 +0.002
Rotation 0.357 0.009 -0.131 0.008 0.064 +0.011
(c, à x)? 0.153 £0.010 -0.110 x0.010 -0.173 +0.015
0.154 0.030 0.535 50.019 0.349 20.041
RMS (m) 0.456 0.456 0.403
Table 2. Surface matching solutions for different heights
Flying height 270m | 450m | 600m
o (conventional) 0.401 0.438 0.496
Mean (conventional) 0.054 -0.062 | -0.104
RMS (conventional) 0.391 0.433 0.496
o (matching) 0.331 0.404 0.414
Mean (matching 0.015 | -0.007 | 0.011
RMS (matching 0.319 0.396 0.405
Number of targets 14 24 22
Table 3. Z difference statistics between conventional and
surface matching orientations (m)
Once the three DEMs were matched, the original surfaces were
transformed by the matching parameters, resulting in the DEMs
being in the same coordinate system as the GPS models.
Comparison with the ground check targets was again possible,
s
id SEE MORE CES
MEC