By Michael S. Loose, Pennsylvania Department of
Transportation, Photogrammetry and Surveys Section (www.dot.state.pa.us),
New Cumberland, Pa., and Michael Shillenn, Photo Science (www.photoscience.com),
Exton, Pa.
In early 2006, the Pennsylvania Department of
Transportation teamed with Photo Science to study the impact of
integrating digital imagery into a traditional large-scale
photogrammetric workflow. Accuracy is always the primary concern when
modifying a large-scale photogrammetric workflow process, followed
closely by cost concerns. There’s always a balancing act between flying
height and accuracy requirements. We wanted to know whether many
long-standing assumptions about standard flying heights would change
when digital imagery became part of the equation.
Study Specifics
To test the comparative accuracy of film vs. digital imagery, a common
site was flown; an aero-triangulation (AT) adjustment, including 14
ground control points, was applied; and map data were extracted. Map
data then were tested against a ground survey of higher accuracy. We
tested map data instead of AT results directly because we anticipated
the digital imagery’s higher quality might make edge detection and image
interpretation more accurate, thus affecting final map accuracy.
However, it was agreed that as a final step we would test the stereo
models resulting from the AT directly.
A project site in Frackville, Pa., was selected. Existing 1:2,400 aerial
photography flown by PennDOT’s Photogrammetry and Survey Section on
April 9, 2005, covered rugged terrain to the north as well as the
relatively flat town of Frackville to the south. This combination of
terrain, coupled with the existing maps, made the site a good choice.
The site was re-flown with a Digital Mapping Camera (DMC) from Zeiss/Intergraph
(www.intergraph.com)
that Photo Science operates from its twin-engine turbine Commanders and
an RC30 film camera from Leica Geosystems (www.leica-geosystems.com)
that PennDOT operates from a Piper Navajo. This time the site was flown
at a scale of 1:4,000 with both sensors. The DMC flight occurred on
April 21, 2005, and the RC30 re-flight occurred on Nov. 18, 2005. The
original 1:2,400 photography acted as a control set, representing a
normal PennDOT project. The 1:4.000 film and DMC flights provided a
direct comparison between film and digital, but also allowed us to
compare the digital imagery with two film sets at different scales.
The existing project control was combined with
new ground control to expand the project area. AT then was performed
on all three sets of photography. Photo Science performed AT on the
digital imagery and the 1:4,000 film. PennDOT performed AT on the
1:2,400 film imagery. All AT was performed according to PennDOT
Publication 122M specifications.
Next, 41 vertical test points and 20 horizontal test points were
selected throughout the map compilation areas. The vertical points
were measured by differential leveling methods, and the horizontal
points were measured using fast-static Global Positioning System
(GPS) methods. All points could be identified on the photography.
Data Collection
Several map compilation areas were selected throughout the project
area, encompassing a mix of terrain from steep to flat, from
forested to heavily developed. PennDOT collected data on the 1:2,400
film imagery. Photo Science collected data on the 1:4,000 film and
1:4,000 digital imagery.
Areas surrounding the surveyed map testing points were identified
with polygons. Photogrammetrists involved in the study were asked to
collect planimetric data in these areas and to create an accurate,
dense digital terrain model. None of the participants knew where the
field survey map test points were located, and they extracted the
data from the stereo models not knowing which data would be tested.
We believed this would add a level of practicality to the test and
introduce factors such as image quality to the final map test
results. If the image edges were crisper, data collection might be
more accurate.
In the direct comparison done by Photo Science, Brian Tolley and
Stephanie Sturtevant split data compilation duties so that each saw
a mix of terrain on both the digital imagery and film imagery. This
allowed for a later discussion regarding hard-to-measure factors
such as image quality, image interpretation, etc.
Test Details
Finally, two separate map tests were performed. First, horizontal
coordinates were taken from the compiled planimetric data and
vertical coordinates were taken from the compiled digital terrain
model. These coordinates were compared to the field survey
coordinates.
Second, the test points were revealed to the operators and the
coordinates were directly lifted from the stereo models. This meant
that operators would find the photo-identifiable points in the
stereo models, position the floating mark over the point as
accurately as possible and record the resulting coordinates.
Coordinates from the field survey weren’t supplied and didn’t bias
the results.
Comments from Participants
Everyone involved in the project agreed that digital imagery was of
significantly better quality than either set of film imagery. All
participants agreed that even the 1:2,400 film wasn’t as good as the
1:4,000 digital imagery. Some of the comments from participants were
as follows:
“Painted lines that were blurry and faded in the film imagery were
crisp with clear edges in the digital imagery.”
“Subtle changes in pavement color easily seen in digital imagery
were almost totally obscured in the film.”
“The ability to zoom in to 4:1 on digital imagery was a big plus.”
“Shadowed areas 100 percent obscured in film imagery were easily
compiled in the digital imagery.”
It should be noted that that the 1:4,000 film imagery was flown in
the fall when conditions weren’t optimal. However, the 1:2,400 film
imagery was flown just a few days prior to the digital imagery and
received similar comments.
The final comment above was mentioned so often
by those involved in the study that it’s worth further discussion.
Because of the dense tree coverage in some areas, steep cliff sides
in others and building lean in the town of Frackville, many areas
were so shadowed that they couldn’t be collected from the film
imagery. However, because of the unique capabilities of the digital
imagery, the same areas were compiled easily. The 12-bit capability
of the DMC is one factor. Another is the fact that film must be
adjusted for overall contrast and brightness throughout the entire
image. By the time the film is scanned, much of the image data
already have been lost during processing. In a purely digital image,
the data are always retained and can be reprocessed for contrast and
brightness continuously throughout use.
Much of PennDOT’s mapping is done in the fall during poor light
conditions and terrain is challenging, so this advantage is
especially noteworthy.
Final Results
Results of the map testing are seen in the charts below. The
digital imagery was better horizontally than either set of
film imagery. This is especially clear in the direct
measurement test in which operators were told where the map
test points were. Without the coordinates—just a description
of the map test locations—operators were able to measure the
points to an astonishing accuracy.
We also concluded from these results that the DMC’s vertical
accuracy is comparable to film flown at the same altitude and
scale. This is also evident from the results reported in the
charts.
We concluded that the transition to digital imagery for
large-scale mapping will be mostly positive. We didn’t factor
in the price of digital sensors, nor did we consider any other
commercially available sensors (see “Digital Airborne Cameras”
chart by clicking on the image at right). Not every digital
sensor is appropriate for large-scale photogrammetry.
What we can say is that current assumptions
regarding flying height, at least for our organizations, won’t
change. Maintaining current flying height specifications for digital
imagery will result in a more accurate map horizontally while
achieving about the same vertical accuracy. Overall, digital imagery
has a positive effect on map accuracy.
Another noteworthy item of the study is that the same ground control
layout was used for all sets of photography. Because the frame size
of the DMC is much smaller, especially longitudinally, the control
density based on number of frames spanned was greater. It took 21
frames from the DMC to cover the same area as 12 frames from the
RC30. If we maintain current ground control layout specifications,
based on the number of frames spanned, we could increase the number
of frames between control. The number of ground control points
required won’t change, just the number of frames between points. It
might be interesting to study the effects of maintaining our current
ground control layout for digital imagery. With ground control every
3 to 4 frames on the DMC imagery, we would expect it to outperform
film at the same flying height.
Photo Science has had a positive experience with its digital
imagery, and since the study the company has purchased a second DMC.
PennDOT believes the transition to digital imagery will be
advantageous in large-scale mapping operations.
Publisher’s Note: Portions of this article were excerpted from a
research paper that originally was presented at the 2006 Annual
ASPRS conference in Reno, Nev.
