of such points will be unique to each survey, recommendations
are provided in Annex C of
ASPRS
2015. The optimal distribu-
tion of horizontal and vertical checkpoints for
UAS
surveys is
still the subject of ongoing research, however, and future revi-
sions of
ASPRS
2015 are likely provide more guidance on check
point configurations appropriate to such surveys
.
ASPRS
2015 also recognizes that data may contain system-
atic biases, which can be estimated by calculating the mean
errors in X-, Y-, and Z-. It is assumed that such biases will be
removed during the processing phase, and therefore do not
have a significant effect on the final accuracy as defined by
RMSE
x
,
RMSE
y
, and
RMSE
z
. Nonetheless
ASPRS
2015 recommends
that mean errors should not exceed 25 percent of the maxi-
mum
RMSE
allowed for a specific accuracy class. Mean errors
exceeding this limit may be permitted under certain circum-
stances, but should be investigated where possible
.
The changes introduced in
ASPRS
2015 also remove the
association between pixel size and horizontal accuracy
contained in the December 2013 draft. This is in recognition
of the flexibility of modern survey methods, whereby similar
pixel resolutions may be achieved by multiple technologies,
with potentially different accuracy implications. For example,
a precisely-calibrated aerial survey camera flown at medium
altitude can deliver the same resolution as a low altitude
UAS
survey carried out using a poorly-calibrated consumer
camera.
ASPRS
2015 includes guidelines for appropriate pixel
size, but stresses that these are only appropriate to orthoimage
mosaics produced using calibrated large and medium format
survey cameras. It is further emphasized that these guidelines
are only intended to be applicable during the transition to
the new standard. For digital orthophotos, the new standards
stipulate that the base resolution of derived orthoimage mosa-
ics should not be less than 95 percent of the original resolu-
tion of the source photos
.
Although
ASPRS
2015 is designed to be independent of
scale, it still allows appropriate horizontal errors to be cal-
culated for maps compiled to specific scales. To do this, the
higher of
RMSE
x
or
RMSE
y
(in cm) is multiplied by a factor of
40 to calculate the equivalent map scale factor (
MSF
) for Class
1 mapping, as defined by
ASPRS
1990. Thus, for a survey with
an
RMSE
x
of 3 cm and an
RMSE
y
of 5 cm, the
MSF
would be 200,
meaning that the survey would meet
ASPRS
1990 Class 1 re-
quirements for horizontal mapping at a scale of 1:200, or Class
2 requirements for horizontal mapping at a scale of 1:100
.
Similarly the
RMSE
z
may be used to calculate the appropri-
ate contour interval as defined by
ASPRS
1990. In this case, the
appropriate interval for Class 1 mapping is found by multi-
plying
RMSE
z
by a factor of three, which for an
RMSE
z
of 10 cm
would give a recommended contour interval of 30 cm. The
recommended contour interval for Class 2 mapping is half
that of the Class 1 contour interval, giving a recommended
contour interval of 15 cm.
To meet the requirements for a specific accuracy classifica-
tion,
ASPRS
2015 specifies that vertical only control points,
and combined horizontal and vertical control points must be
surveyed to a higher standard of accuracy, which should meet
the following requirements:
RMSE
x
,
RMSE
y
or
RMSE
z
= ¼ *
RMSE
x(Map)
,
RMSE
y(Map)
or
RMSE
z(DEM)
(3)
where
RMSE
x(Map)
and
RMSE
y(Map)
are defined by the appropriate
horizontal accuracy class, and
RMSE
z(
DEM
)
is defined by the ap-
propriate vertical accuracy class. For horizontal only control,
the corresponding requirements are:
RMSE
x
or
RMSE
y
= ¼ *
RMSE
x(Map)
or
RMSE
y(Map)
and
(4)
RMSE
z
= ½ *
RMSE
x(Map)
or
RMSE
y(Map)
.
Case Studies
Methodology
Aerial surveys were performed at two sites in 2013 and 2014.
The first flight was carried out in late-September 2013 along
the Evan Thomas Creek in Kananaskis Country of southern
Alberta. This survey was carried out following a devastating
flood, which occurred in June 2013. The second flight was
carried out in June 2014 along the Sunwapta River, approxi-
mately 3.5 km downstream of the Athabasca Glacier in Jasper
National Park, Alberta. Both sites are characterized by broad
exposures of river gravel bordered by vegetated terrain. The
main focus of the surveys and measurements were the areas
devoid of vegetation.
DEM
s and orthoimage mosaics generated
from each survey are shown in Figure 1
.
The
UAS
used in the surveys was the eBee (Sensefly, Ltd.)
fixed-wing platform. The eBee is both lightweight (700 g) and
small (0.96 m wingspan), and acquires geotagged images with
a consumer-grade, 16.1 megapixel Canon IXUS 127 HS digital
camera. The eBee is hand-launched and flies autonomously
along pre-defined flight lines established through proprietary
flight planning software. We set the image overlap at 80 per-
cent, with 70 present sidelap across flight paths for both sur-
veys. The eBee was flown at a height of 122 m above ground
level, resulting in five flight lines and 188 images at Evan
Thomas and eight flight lines and 305 images at Sunwapta.
GCP
s were established at each site prior to the
UAS
surveys: a
total of 22 at Evan Thomas and 20 at Sunwapta. The
GCP
s con-
sisted of painted X’s clearly visible in the imagery. A Trimble
R8
GNSS
(
GPS
+
GLONASS
) with real-time kinematic corrections
was used to measure the center coordinates of each
GCP
. In
both projects the 3D accuracy of each
GNSS
point was
≤
0.02 m
RMSE
(set as the maximum allowable threshold in the unit). A
higher accuracy threshold of 0.01 m
RMSE
was attempted, but
was not achievable due to the mountainous terrain. The
GNSS
was also used to measure the coordinates of horizontal and
vertical checkpoints distributed at each site. At Evan Thomas,
33 horizontal checkpoints were measured, comprising the
centers or pointed edges of well-defined features (boulders,
logs, etc.). During the field survey we acquired photographs of
the exact measurement point on these features and used these
to pinpoint the location on the orthoimagery. At Sunwapta
we used 30 painted X’s and recorded the center coordinates.
We also acquired two large datasets of vertical checkpoints (
n
= 167 at Evan Thomas,
n
= 172 at Sunwapta). These were not
marked, and were acquired in locally-flat non-vegetated areas.
As with the
GCP
s, horizontal and vertical checkpoints were ac-
quired by
GNSS
, with a 3D accuracy threshold of
≤
0.02 m
RMSE
.
All
GCP
s, as well as horizontal and vertical checkpoints used
in both surveys are shown in Figure 1.
Processing was carried out using Pix4D SfM software.
Pix4D does not require an input camera calibration, since the
calibration is computed as part of the process, and as such it
is recomputed for each survey, thereby avoiding many of the
problems associated with the instability of consumer cameras.
The software automates traditional photogrammetric steps
and incorporates the
SIFT
algorithm (Lowe, 2004). Preliminary
estimates of camera positions derived from the onboard
GPS
are used as a first approximation. The relationship between
the images is then established by matching key features be-
tween multiple images. The surveyed horizontal and vertical
GCP
s are used to complete the
EO
process, with the final
EOP
being determined through a rigorous least squares adjust-
ment. After the
EO
process is complete, a 3D point cloud can
be generated, from which a
DEM
can be interpolated. This
DEM
is then used to orthorectify the individual images. The final
step is to combine the orthorectified images to form a seam-
less mosaic. Orthoimage mosaics and
DEM
s from both surveys
were produced at a spatial resolution of 3.5 cm (Figure 1).
790
October 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
