Professional Documents
Culture Documents
High-Resolution Mapping with Unmanned Aerial Systems
High-Resolution Mapping with Unmanned Aerial Systems
Surveying and Land Information Science, Vol. 74, No. 1, 2015, pp. 5-13
camera), a laptop computer which is used for flight Project Preparation
planning and for monitoring the performance of
the UAV, and a remote control (RC) transmitter The boundary of the mapping area should be
which can be used to manually control the UAV defined [Google Earth (GE) can be used to do
and monitor the telemetry and battery status. this] prior to any other planning and a decision
made on required resolution, typically expressed
in terms of ground sampling distance (GSD). Once
Unmanned Aerial Vehicles these have been specified several other preparatory
steps should be taken, as summarized below:
UAVs can be divided into two general categories—
copters (known as VTOL) and fixed wing vehicles. Field Testing
VTOL vehicles are designed as multirotor copters It is crucial that the entire UAS be tested and
with 3, 4 (“quadcopter”), 6 (“hexacopter”), or evaluated prior to embarking on a project. In
8 propellers (“octocopter”). We used a quadcopter our case we set out some 50 control points at an
for the pilot tests described here and almost all of easily accessible site and accurately surveyed
our experience has been with these types of UAVs. these with a dual frequency global navigation sat-
Fixed wing UAVs are designed and fly much ellite system (GNSS). This resulted in coordinates
like conventional aircraft. Unlike VTOLs, they determined to within an accuracy of 1-2 cm. This
require a significant area to take-off and land and control site was then flown, imagery acquired at
must maintain forward motion to stay in the air. different flying heights, and resulting data pro-
VTOL vehicles, as the name implies, can take off cessed using Multiview Stereopsis (MvS) software
vertically and therefore require very little space (in our case we used Photoscan from Agisoft). A
for take-off and landing and have the ability to small number of the control points were used as
hover if necessary. VTOLs are therefore preferable ground control for the processing and the remain-
for smaller areas where high-resolution imagery is ing points served as check points to evaluate the
required. Fixed wing UAVs are more productive spatial quality of the resulting orthophotos.
over larger areas, although the imagery will gener-
ally be at a lower resolution because of the higher
Regulatory Limitations
speeds required to stay aloft as well as the restric-
The regulatory requirements should be researched
tions on camera shutter speeds and continuous
before any decisions are made on the equipment
exposure rates. When selecting a UAV it is impor-
and operational approach for the project. The reg-
tant to consider whether coverage of a whole area
ulatory requirements may depend on the take-off
is needed at the same time, or whether mapping
weight of the UAV. In Germany, for example,
incrementally on an as-needs basis is better. The
UAVs heavier than 25 kg are prohibited, but
advantage of the latter approach is that it pro-
general authorization can be obtained to fly
duces more current spatial data which can be an
UAVs with take-off weights of less than 5 kg. In
issue in rapidly developing areas, such as informal
other countries, like the U.S., the federal govern-
settlements in developing countries.
ment through the FAA has taken a heavy-handed
regulatory stance by prohibiting all commercial
UAS flights until such time as they establish regu-
UAS Mapping Methodology lations. In 2008, the FAA issued guidelines enti-
tled “Interim Operational Approval Guidance”
The sequence of tasks involved in UASs mapping (FAA 2008) which outlined the process for apply-
is summarized in Figure 1 and described more ing for a Certificate of Authorization (COA).
fully in the following sections. This authorization is only available to public enti-
ties. Generally, model airplanes are
restricted to flying below 400 ft,
more than 3 miles from an airport,
and away from highly populated
areas (FAA 2014). The FAA is sup-
posed to finalize regulations for the
operation of UASs in the U.S. by the
end of 2015.
However, a judge with the National
Figure 1. UAS mapping procedure. Transportation Safety Board ruled
+
has fewer restrictions. When traveling on com- and (4) the required forward and lateral overlap.
mercial airlines it is much more convenient to In our pilot tests at Albania, we used a 20 Mega-
use a UAV that can be fitted into carry-on lug- pixel camera with a 16 mm focal length and a
gage. Larger UAVs require prior shipping by air- 23.5 15.6 mm sensor with a pixel size of 4.8 mm.
+
freight, which increases the logistical complexity Flying at a height of 75 m above the ground gave
and the risk of delays in customs. Very strict us a GSD of 18 mm in the agricultural test area.
packaging and shipping specifications apply to In the urban test area, we flew 50 m above the
lithium-polymer (LiPo) batteries which should be rooftops using the same camera giving us GSDs
transported as carry-on luggage in a fire-proof case. varying from 13 mm on the rooftops to 20 mm at
ground level.
Flight planning was done with Mission Planner,
Reconnaissance an open source software package (which can be
Even though GE provides a good initial view of downloaded at http://ardupilot.com/downloads/
the project area, an on-site reconnaissance is ?did=82). Flight plans can be edited through
essential to identify dangerous obstructions such either a graphical or tabular interface (see
as cell towers, power lines, or other objects not Figure 2). Flight plans can be rendered in GE
visible on the GE imagery. The reconnaissance is to verify intended waypoint coordinates and to
Figure 2. Mission planner interface showing flight lines (yellow) for one flight.
datasets (each photograph varies from 4 to 5 mega- network (TIN), and texture atlas as shown below
bytes in size). In Albania, for example, to cover in Figure 5.
an area of approximately 24 hectares (58 acres) Once the original images are uploaded, the
in size at a flying height of 75 m required processing software reads the basic camera geom-
492 overlapping aerial photos amounting to etry (focal length and sensor dimensions) from
2.3 gigabytes of data storage space. Because of the EXIF header of the image files and performs
this huge volume of data, the image processing a photo alignment. This produces relative camera
is usually the most time-consuming part of the exposure coordinates, a camera calibration for
UAS work flow. each and every image as well as a sparse point
Because nonmetric cameras are not geometri- cloud. For low accuracy requirements this may be
cally stable (a basic requirement for classic photo- all that is required to build a model if the final
grammetric mapping), the imagery is processed product does not have to be referenced to some
using a combination of classical photogrammetry spatial reference frame. However, the process of
and MvS. Photoscan, a Russian program by determining the CEPs is significantly accelerated
Agisoft that employs the MvS algorithm, follows if approximate positions are introduced into
a fairly simple work flow which includes: the photo alignment procedure. Typically, these
importing aerial imagery, photo alignment, approximate positions are obtained from the
introducing ground control (if necessary), build- stand-alone navigation GPS receiver mounted on
ing models of the terrain in the form of sparse board the UAV. However, if the product is to be
and dense point clouds, triangulated irregular accurately geo-referenced to a defined coordinate
approximately 65 hectares. This site was chosen it easy for landholders and their family members
to test the performance of the UAS in high to identify their parcel boundaries. The resulting
density urban environments. parcel boundaries for a section of the coverage
A strip of the Elbasan national highway (800 m area are shown in Figure 8 below.
long) approximately 10 km from Tirana that Although Albania is fortunate enough to have
had been partially constructed. This was selected reasonably high-resolution (GSD = 20 cm) ortho-
to investigate the application of UAS for moni- photos for that area, these are based on photogra-
toring and managing linear features such as phy taken in 2007 and a number of changes have
roads, powerlines or pipelines. taken place since that time. This can be clearly seen
A flying height of 75 m above the ground was in the comparison of a small portion of our UAS-
used for Fushe Milot and the Elbasan Highway, derived orthophoto with the same area in the 2007
while at the Komuna Farke site we flew at a height orthophoto (see Figure 9).
of 50 m above the tops of the buildings. In all, We tested the spatial quality of the UAS-derived
we flew a total of 21 flights covering a distance of orthophoto by exporting it into ArcGIS and mea-
about 40 km and acquiring some 4200 aerial photo- suring the coordinates of all GCPs. Since we only
graphs with a ground resolution of approximately used 9 out of the 52 GCPs for geo-referencing
2 cm. the imagery, the remaining control points can
We were able to produce DEMs, orthophotos be regarded as independent checks on the
and 3D models for each of these areas. In the orthophoto. Table 1 below summarizes the com-
case of the agricultural area of Fushe Milot we parison of the resulting coordinates with those
returned to the area a week after we flew the that we measured using kinematic GPS (regarded
photography for that area and conducted a rapid as known).
boundary adjudication of approximately 29 par- While these statistics are reasonably good and
cels. The high resolution of the current spatial consistent with other tests (Eisenbeiss et al. 2013),
data shown in the orthophoto (see Figure 7) made we are concerned about a handful of GCPs
Figure 9. Portion of UAS-derived orthophoto (left) compared with 2007 orthophoto (right).
which had positional errors between 10 and We are currently conducting additional analyses
14 cm. Other studies have managed to achieve of these data to identify possible sources of
“precisions” of 2-3 cm in the horizontal and 4- error and to improve the 3D modeling of com-
10 cm in the vertical (Devriendt 2014, p. 23). plex landscapes.