498
June 2014
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
Dr. Abdullah:
You are right; according to the draft of the new standard
Class1 (which is recently renamed to “Engineering Class”)
aerial triangulation accuracy should meet ½ pixel or 1.25 cm
in your case, see Table 1.
To achieve such accuracy, the following conditions need to be
met:
1)
Ground control accuracy should be in the order of ¼
pixel or 0.625 cm for this project. This may require
conventional surveying techniques, as a GPS-based
survey may not meet such stringent requirements.
2)
For this project, the estimated accuracy or the apriori
weight used for the ground control of 2-3 cm (0.71
-1.2 pixel) is a little loose if you expect the aerial
triangulation to meet an accuracy of ½ pixel (or 1.25
cm) of the final ortho products. By tightening the
apriori accuracy estimated value for the GCP from 2-3
cm (0.71 -1.2 pixel) to about ¼ - ½ pixel (or 0.625 cm
to 1.25 cm) during the bundle block adjustment, you
may notice improvement in the GCPs and checkpoints’
fit. Here, unless the actual accuracy of the ground
controls supports such stringent apriori weight, one
should expect a false or misleading improvement
in the resulted accuracy. False constrains on the
ground controls may cause the errors in the GCPs
and checkpoints to be pushed around to the image
measurements or the airborne GPS and IMU-based
exterior orientation parameters. The last statement
is especially true when you consider that the apriori
weight of the airborne GPS (ABGPS) position was
set to be equal to 5 cm, 5 cm and 10 cm in X, Y and Z
respectively, which is relatively loose compared to the
given GCPs accuracy.
To meet the Engineering Class of the new standard for this
project, I recommend one or both of the following:
1)
Assuming that the accuracy of the ground-control
network is within ¼ pixel (RMSE), try my suggestion
of changing the weights as I described in (2) above and
run the solution again.
2)
Since you have enough ground controls, run a
conventional aerial triangulation (without the use of
airborne GPS or IMU) using most of the GCPs in the
solution but leaving no more than 30 well-distributed
checkpoints. You need to remember to change the weight
on the GCPs as described in
(2) above. This solution will
most likely fit the quality of
the ground control points for
the project (good or bad), as it
excludes from the solution the
camera positions and orientation
constraints as determined by
the ABGPS and the IMU. Such
an approach is very realistic,
and I am usually hesitant of
using ABGPS and IMU in an
aerial triangulation with such
stringent accuracy requirements.
Finally, the newly introduced digital sensors and the new
low-flying platforms such as UAS are resulting in extra-fine
ground resolution or GSD, which sometimes makes it difficult
if not impossible to meet the Engineering Class (Class I) of
the new proposed accuracy standard without extra efforts.
UAS, for example, is providing imagery with a GSD of 1.5 cm,
which translates to aerial triangulation accuracy requirement
of 0.75 cm and ground control accuracy requirement of 0.37
cm (that is only 0.148 inch!) according to the Engineering
Class. Such figures present a great challenge for the current
GPS-based surveying techniques and the photogrammetric
process. A new paradigm needs to replace the old one in
our thinking, and in order to adjust for the more accurate
and modern map accuracy standard, not all maps should
be required to meet the Engineering Class (Class I). There
is nothing wrong in saying that such extra-fine resolution
products need to meet the Mapping Class (Class II) or even
the Planning Class (Class III). Mapping Class accuracy for
1.5 cm ortho is still only 3.0 cm (or 1.18 inch), which is still
awfully accurate for many applications. For applications that
require higher accuracy (i.e. Engineering Class) and very
high-resolution imagery, very accurate ground controls that
most likely will need traditional surveying techniques such as
total station and leveling, quality metric camera and rigorous
photogrammetric workflow may be needed.
**Dr. Abdullah is Senior Geospatial Scientist at Woolpert,
Inc. He is the 2010 recipient of the ASPRS Photogrammetric
(Fairchild) Award.
The contents of this column reflect the views of the author,
who is responsible for the facts and accuracy of the data pre-
sented herein. The contents do not necessarily reflect the offi-
cial views or policies of the American Society for Photogram-
metry and Remote Sensing and/or Woolpert, Inc.
Table 1 Horizontal accuracy classes for orthophoto according to the new standard
Accuracy Class
RMSE
x
and RMSE
y
Aerial Triangulation or
INS-based RMSE
x
RMSE
y
and RMSE
z
Ground Controls
Accuracy RMSE
x
RMSE
y
and RMSE
z
Engineering
Pixel size x 1.0
Pixel size x 0.5
Pixel size x 0.25
Mapping
Pixel size x 2.0
Pixel size x 1.0
Pixel size x 0.50
Planning
Pixel size x 3.0
Pixel size x 1.5
Pixel size x 0.75
IV
Pixel size x 4.0
Pixel size x 2.0
Pixel size x 1.00
…
N
Pixel size x N
Pixel size x 0.5N
Pixel size x 0.25N
