PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2014
497
I recently attended the panel discussions during the
ASPRS 2014 Annual Conference in Louisville regarding
the draft for the “ASPRS Positional Accuracy Standards
for Digital Geospatial Data.” In general, I really like the
organization and applaud the switch to metric stan-
dards. What I really appreciate is the simple grouping
of accuracy that purchasing organizations can use
according to their application with high clarity. As far
as I am concerned, we should just adopt this type of
standard in Canada.
However, there are a few reservations I have about
proposed specifications. Consider the following aerial
triangulation specifications, which we executed
recently:
This project is a good modern AT example for high
accuracy. The client, who is an engineering/ photogram-
metry firm, was mapping a light-rail corridor with aerial
photography to replace an engineering survey with a
total station. As such, they established enough control
to do “special-order” work.
We assessed the accuracy in this case by setting every
other point as check (which should be a very good
representation of the “true” accuracy of the system).
Because of the level and quality of the control that is
being used (and the state-of-the-art airborne equip-
ment), this should be Class I work.
How I have been assessing the aerial triangulation
accuracy actually stems from a mapping matters article
which reported the accuracy as a function of pixel size
and control and checkpoint residuals. I checked the
result of many digital photogrammetry results and also
those of film (scanned at 14 um, which seems to be the
accepted ‘sweet spot’ for film achievable accuracies)
and the traditional 30 um at plate-scale formula. The
pixel-based accuracy standard seems very
repeatable as well as logical, which is why
I like this draft standard.
What I have repeatedly obtained from
aerial triangulation solutions is an RMSE of
~0.6 to 0.7 of a pixel on fixed ground control
points and ~1.2 to 1.3 pixel on checkpoints.
We use the rule that the mapping product
would be double that, being ~2.4 to 2.6
pixels.
In any event, the results are what I think
are close to well-functioning AT; however,
this AT wouldn’t meet Class I of the pro-
posed ASPRS standard. I think that the
standard thresholds are slightly too tight.
Could you perhaps give guidance of how
the results may be interpreted differently?
Robin Poot
Geomatics Manager
The Airborne Sensing Corporation, Toronto, Ontario
CANADA
Parameter
Value
Comments
GSD
2.5 centimetres
GPS/ IMU
Applanix 510
Camera
Vexcel Ultracam X
Image Weight
2 um
Assumed GCP
accuracy
2 and 3 cm xy and z
Normalized to component
variance ~= 1.0
Assumed GPS
accuracy
5 and 10 cm xy and Z Loose, yes, but I find this
is ok
Number of photos
485
Number of ground
control points
462
Obviously a lot more than
is normally required
Resulted GCPs fit
RMSE
x
= 2.4 cm (0.96 pixel)
RMSE
y
= 2.1 cm (0.84 pixel)
RMSE
z
= 1.9 cm (0.76 pixel)
Resulted check
points fit
RMSE
x
= 2.8 cm (1.12 pixel)
RMSE
y
= 3.2 cm (1.28 pixel)
RMSE
z
= 3.7 cm (1.48 pixel)
