Id 4/-
acility
Nhich
lation
) one
the x
;ause
r the
focal
from
3 mm
| and
g the
ing.
diting
] and
ery is
any
prior
( into
owlett
an be
) dye
nding
ts in
1 the
castle
1g an
5m).
ith its
; was
strip.
ise t0
Would
RMS
when
ssary
] the
nt for
lentify
itable
(and
n, the
; with
easily
were
1d.
'imble
annel
jer for
eight,
back-packed design makes single man surveying feasible
and it is ideally suited to the collection of
photogrammetric control points. Using post-processed
differential correction, Trimble claim the corrected C/A
code provides an accuracy of ‘better than one metre’ on a
second-by-second basis. There are however several
degrading factors to this, such as the distance to the
base station and the number of satellites available. The
survey was carefully planned to avoid times of poor
satellite configuration and the Surveying Department's
SciNet pillar was used as a base station in the centre of
the survey area so keeping the distance between base
and rover below 1 km. The main degrading factor (other
than selective availability) of the survey in this case was
found to be that of multipathing - a problem that is very
difficult to overcome in a city centre site when using only
C/A code differential processing. In an attempt to
surmount this problem, each control point was observed
with the antenna mounted on a bipod for a period of five
minutes so as to provide an averaged position.
The points were differentially post-processed and
converted from WGS84 to OSGB36 using a four
parameter local transformation. ^ This allowed the
referencing of the imagery to the OS data of the area. A
check on a point of known co-ordinates at the periphery
of the survey area showed this method of GPS
observation to give average errors of 0.84 m in plan and
1.25 m in height. Unfortunately this is larger than the
ground pixel size, although the collection of some 60
control points allowed rejection of any inaccurately co-
ordinated points. The two images being used for the
survey were orientated and rectified using the GCPs.
The space resection performed to orientate the imagery
yielded the orientation parameters shown in Table 1. It
can be seen from the GCP residuals that the RMS error
is at the sub-pixel level.
Image 81 Image 82
X (m) 424784.37 424634.54
Y (m) 565100.22 564858.91
Z (m) 1494.56 1494.43
degrees) -4.160 -2.993
degrees) 3.305 0.689
K (degrees) 151.605 151.149
Datum height (m) 58.46 64.28
Base/Height ratio 0.19
No. of GCPs 36 43
RMSE (pixels) 0.96 0.86
RMSE (m) 0.44 0.39
posterizing effect and setting the measuring mark can
prove difficult in areas of low contrast.
The stereopair was then autocorrelated to produce a
Digital Elevation Model (DEM) of the area. This enabled
the imagery to be geometrically corrected for relief
displacement to create an orthophoto. The area based
matching technique employed by the DMS is suited
ideally to gently undulating areas and inevitably the
disjointed nature of a city centre site causes many
failures in correlation, leading to lengthy editing times.
Every pixel in the image was correlated and rectified in
an attempt to correct the buildings for lean. There is
considerable masking of ground detail due to building
lean on the imagery, and a facility to recover this from
adjacent images, such as the technique employed by the
Leica Helava system (Simmons, 1996) would be useful in
such circumstances. Measurements on the orthophoto of
the same GPS control points used in the stereo tests
produced RMS errors of 0.94 m (2.1 pixels) in plan and
1.71 m (3.8 pixels) in height.
3.4 Mapping Accuracy
Measurements taken against the GCPs are unlikely to
give a true representation of the absolute accuracy of the
survey since they have been used in orientating the
imagery. To ascertain the absolute accuracy of co-
ordinates produced from the images, a total of 70
randomly chosen points were measured from the OS
digital mapping of the area. This data has a quoted RMS
error of 0.4 m (Ordnance Survey, 1995). The same
points were then measured on the stereomodel and the
orthophoto (with heights from the DEM). The results of
the comparison can be seen in Table 2. Unfortunately
the table shows no measure of heighting accuracy or
precision since the OS dataset is two dimensional. The
stereomodel and orthophoto measurements were
therefore compared, yielding an RMS error in height
between the two of 0.90 m (2 pixels). Precision (obtained
by 10 repeated measurements to 6 different points) for
the heighting was calculated as 0.39 m (0.9 pixels) RMS
error for the stereo measurement. Orthophoto height
measurements showed almost perfect repeatability,
although this can be attributed to the resolution of the
DEM.
Table 1: Orientation parameters and GCP residuals of
the imagery used in the Newcastle survey.
3.3 Photogrammetric Processing
Stereoscopic tests on the 15 GCPs on the overlap
produced RMS errors of 1.02 m (2.3 pixels) in plan and
1.79 m (4.0 pixels) in height. The accuracy to which one
Can measure is, however, degraded by the DMS’s
anaglyph stereoviewing facility. Despite the fact that the
monitor can display 32,000 colours, the display driver
only supports 8 bit colour with each of the images
displayed using four bits (16 grey levels) to represent the
296 levels recorded by the camera. This gives a
565
Accuracy
Method X RMSE Y RMSE
Metres Pixels Metres Pixels
Stereo 0.79 1.8 0.55 1.2
Ortho 0.52 1.2 0.48 1.1
Repeatability (Precision)
Method X RMSE Y RMSE
Metres Pixels Metres Pixels
Stereo 0.09 0.2 0.10 0.2
Ortho 0.10 0.2 0.09 0.2
Table 2: Accuracy and repeatability statistics for the
Newcastle survey (comparison against OS).
As a check, the GPS control was also compared to the
OS data, resulting in an RMS error of 0.84 m in the east-
west (x) direction and 1.16 m in the north-south (y)
direction.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
