remote sensing

Article
Optimization of UAV Flight Missions in Steep Terrain
Klemen Kozmus Trajkovski *, Dejan Grigillo and Dušan Petrovič
Faculty of Civil and Geodetic Engineering, University of Ljubljana, Jamova 2, SI-1000 Ljubljana, Slovenia;
dejan.grigillo@fgg.uni-lj.si (D.G.); dusan.petrovic@fgg.uni-lj.si (D.P.)
* Correspondence: klemen.kozmus@fgg.uni-lj.si; Tel.: +386-1-4768-648




Received: 1 March 2020; Accepted: 17 April 2020; Published: 19 April 2020


Abstract: Unmanned aerial vehicle (UAV) photogrammetry is one of the most effective methods for
capturing a terrain in smaller areas. Capturing a steep terrain is more complex than capturing a flat
terrain. To fly a mission in steep rugged terrain, a ground control station with a terrain following
mode is required, and a quality digital elevation model (DEM) of the terrain is needed. The methods
and results of capturing such terrain were analyzed as part of the Belca rockfall surveys. In addition
to the national digital terrain model (NDTM), two customized DEMs were developed to optimize
the photogrammetric survey of the steep terrain with oblique images. Flight heights and slant
distances between camera projection centers and terrain are analyzed in the article. Some issues were
identified and discussed, namely the vertical images in steep slopes and the steady decrease of UAV
heights above ground level (AGL) with the increase of height above take-off (ATO) at 6%-8% rate.
To compensate for the latter issue, the custom DEMs and NDTM were tilted. Based on our experience,
the proposed optimal method for capturing the steep terrain is a combination of vertical and oblique
UAV images.

Keywords: UAV; vertical images; oblique images; steep terrain; rockfall; DTM; custom DEM; tilted
DEM; flight mission

1. Introduction
Many mountains in the world consist of sedimentary rocks such as limestone, dolomite, flysch,
conglomerates, sandstone or even combinations of these rocks and molds. Their characteristic are
the layers of coatings, which can be strongly differentiated due to later earth folds. Due to their
lower compactness and porosity, some of them are also cut through by water currents, which further
reduce their stability when large amounts of water and dirt are present. In the steep slopes of this
type of rock, rockfalls and landslides are a common phenomenon. The consequences are scree, decay,
debris flows and terraces of deposited material. Most of these areas are uninhabited and rarely
visited, so that the consequences do not cause any significant damage to people and their possessions.
However, these events damage vegetation and alter natural habitats and ecosystems. On the other
hand, if potentially unstable areas are sufficiently close to populated areas or if the amount of broken or
cracked material is large enough to reach populated areas either in the form of debris flows, rockfalls,
landslides or mudflows, they may cause significant damage to facilities, infrastructure, forests or
agricultural crops. Therefore, all potentially dangerous hinterland areas should be regularly monitored
and timely anticipated for possible events and appropriate measures should be taken to prevent or at
least mitigate possible consequences [1].
Surface monitoring is usually carried out using either point-based techniques or surface-based
techniques. Point-based techniques, such as global navigation satellite systems (GNSS), extensometers,
total stations, laser and radar distance meters, generally offer better precision, but they only provide
information on some selected monitoring points [2,3]. Surface-based techniques (photogrammetry,

Remote Sens. 2020, 12, 1293; doi:10.3390/rs12081293 www.mdpi.com/journal/remotesensing

Remote Sens. 2020, 12, 1293 2 of 20

satellite-based or ground-based radar interferometry and terrestrial or airborne laser scanning) mainly
belong to the field of remote sensing [4] and are capable of monitoring the entire surface. However,
most of these methods are associated with enormous costs and a lack of spatial and temporal resolution
for monitoring most slope deformations [5]. Unmanned aerial vehicle (UAV) based remote sensing
belongs to the domain of surface-based techniques and is a cost-effective alternative to data acquisition
with high spatio-temporal resolution [6]. Although different sensors can be mounted on UAVs,
mapping is generally based on images taken by digital cameras [7]. Together with the development of
image processing techniques such as multiview stereo (MVS) and especially structure from motion
(SfM), UAVs offer effective and cost-efficient photogrammetric techniques to obtain high-resolution
data sets [8].
Several researchers used images from UAVs as a data source to measure changes in steep terrain.
To obtain reliable results from images taken in steep terrain, flight planning of the UAV is a very
important step. The main parameters in flight planning for UAVs are the definition of the area of
interest, selection of the flight altitude above ground level (AGL), flight speed, forward and side
overlap of successive images and parameters of the digital camera (sensor dimension, pixel resolution
and focal length). All these parameters influence the ground sampling distance (GSD) of the images as
well as the accuracy of the final results [7]. A comprehensive overview of mission planning techniques
using passive optical sensors (cameras) is given in [9].
Manconi et al. [7] developed a routine that divides the area of interest into several flight lines
and each of which has a certain height, so that the average distance to the surface is maintained.
The optical axis is perpendicular to the ground. Similar was done in [10]. De Beni et al. [11] surveyed
volcano Mt. Etna with a modified even-height mission, whereby the elevations of the upper waypoints
were raised. The result is that the entire flight plan is on an inclined plane. Niethammer et al. [12]
observed a landslide in France. The average inclination of the area was 25 degrees. They used
a manual control of the flight heights. Möllney and Kremer [13] dealt with the so-called contour flying,
but for manned aircrafts. However, a clear advantage of the less modulated resolution is emphasized.
The authors of [14] used the mission planning for the UAV- based landslide monitoring in Canada,
but at a constant height.
Some papers contain only partial details of the UAV flight. Rossini et al. [15] surveyed a glacier
retreat in the Alps. The adaptation of flight heights is mentioned, but no further specifics about flight
altitudes are given. Similarly, [16] documented a geomorphic change detection of a gorge in Taiwan
with images at different heights, but no flight specifications are revealed. Valkaniotis et al. [17] mapped
an earthquake-related landslide in Greece. Of the actual UAV survey, only vertical and oblique views
are mentioned. Agüera-Vega et al. [18] dealt with an extreme topography in an almost vertical road
cut-slope in Spain using UAV. They took horizontal images in four flight lines on a vertical plane and
oblique images taken at 45 degrees downwards in two passes. Some authors do not mention any
specifics of the UAV survey. UAV was used by [19–21] to survey landslides and rockfalls.
Yang et al. [22] conducted a study on the optimization of flight routes for the reconstruction of digital
terrain models (DTM). The procedure assumes a constant UAV height. Pepe et al. [9] gave an overview
of the airborne mission planning of the current platforms and sensors. The terrain-following planning
mode is only mentioned.
Some of the methods described above improve the suitability of UAV surveys in sloped terrain,
either by tilting the flight plane, defining different heights of the flight lines or by flying the UAV
manually. These solutions are of limited use in very uneven terrain where the elevations of the relief
shapes vary greatly in all directions. The solution would be to adjust the UAV height according to the
terrain elevation each time an image is captured. In our opinion, the terrain-following flight mission is
the only reasonable solution.
Our goal was to design customized digital elevation models (DEMs) that can optimize mission
planning in steep terrain. The custom DEMs modify the actual DTM in a way that allows more
consistent UAV distances from the terrain when used for mission planning. This ensures more consistent
Remote Sens. 2020, 12, 1293 3 of 20

Remote Sens. 2020, 12, x FOR PEER REVIEW 3 of 20

image overlap, more consistent GSD of images and reduces blind spots. These facts contribute to the
contribute to the efficiency of the dense image matching algorithms used to calculate dense point
efficiency of the dense image matching algorithms used to calculate dense point clouds. The custom
clouds. The custom DEM also increases flight safety, as uniform distances of the UAV to the terrain
DEM also increases flight safety, as uniform distances of the UAV to the terrain prevent the vehicle
prevent the vehicle from colliding with exposed terrain points. To support the UAV mission flying,
from colliding with exposed terrain points. To support the UAV mission flying, two custom terrain
two custom terrain models and their alternatives have been developed. Additionally, the capturing
models and their alternatives have been developed. Additionally, the capturing of oblique images is
of oblique images is recommended either instead of or in addition to the conventionally used vertical
recommended either instead of or in addition to the conventionally used vertical images.
images.
2. Methodology
2. Methodology
2.1. UAV Height, Elevation and Altitude
2.1. UAV Height, Elevation and Altitude
Basic vertical distances of UAV flying are shown in Figure 1. Elevation is the vertical distance
Basic vertical
of a ground distances
point above meanof UAV flying
sea level are shown
(AMSL). in Figure
Altitude is the1.vertical
Elevation is the of
distance vertical distance
the current UAVof
a groundAMSL.
position point The
above mean
height sea level
above (AMSL).
take-off (ATO) Altitude is thebetween
is a difference vertical the
distance of and
altitude the current UAV
the elevation
position
of AMSL.
the take-off TheAMSL.
point heightThe above
heighttake-off (ATO)
AGL refers to is
theavertical
difference between
distance the the
between altitude
UAV and the
in flight
elevation of the take-off point AMSL. The height AGL refers to the vertical distance between
and the ground. Another important distance is the distance from the projection centre of the camera to the UAV
in flightalong
ground and the
theground. Another
optical axis. When important distance
the camera is thedown
is pointed distance from
in the thedirection,
nadir projectionthe centre of the
distance is
camera to ground along the
equal to the UAV height AGL. optical axis. When the camera is pointed down in the nadir direction, the
distance is equal to the UAV height AGL.

Figure 1. Unmanned aerial vehicle (UAV) heights, elevation and altitude. UAV's take-off point is
Figure 1. Unmanned aerial vehicle (UAV) heights, elevation and altitude. UAV's take-off point is
colored grey and the current UAV position is black.
colored grey and the current UAV position is black.
For oblique images, the slant distance from the projection centre of the camera to the ground point
For oblique images, the slant distance from the projection centre of the camera to the ground
in the direction of the optical axis is labeled as DCG (distance camera-ground). Figure 2 shows the
point in the direction of the optical axis is labeled as DCG (distance camera-ground). Figure 2 shows
DCGs for vertical and oblique images.
the DCGs for vertical and oblique images.
2.2. UAV Photogrammetry in Flat and Non-flat Areas
2.2. UAV Photogrammetry in Flat and Non-flat Areas
If the terrain is flat or almost flat, the usual method of capturing the terrain with a UAV is to
If the terrain
fly horizontally atisa flat or almost
certain heightflat,
ATO,thewhich
usualrefers
method of capturing
to the the terrain
vertical distance with
above a UAV
the launch is point.
to fly
horizontally at a certain height ATO, which refers to the vertical distance above the launch
The camera of the UAV is pointed in the direction of the nadir. The images are taken with a certain point. The
camera of the UAV is pointed in the direction of the nadir. The images are taken
overlap to allow 3D reconstruction of the ground using image matching algorithms. Numerous with a certain overlap
to allow 3D reconstruction
applications, such as Mission of the ground
Planner, usingDJI
UgCS, image matching
GS Pro, algorithms.
3DSurvey Numerous applications,
Pilot, Pix4Dcapture and others
such as Mission Planner, UgCS, DJI GS Pro, 3DSurvey Pilot, Pix4Dcapture
offer automated mission flying. A flight plan is generated based on a user-defined and others offer
area,automated
the flight
missionoverlap
height, flying. of
A the
flight plan and
images is generated basedThe
camera type. on plan
a user-defined
is uploadedarea, theUAV,
to the flight height,
which thenoverlap of
takes off
the images and camera type. The plan is uploaded to the UAV, which then takes off automatically,
flies to the waypoints, takes images and lands at the home point, provided there are no malfunctions.
Remote Sens. 2020, 12, 1293 4 of 20

automatically, flies to the waypoints, takes images and lands at the home point, provided there are
noRemote
malfunctions.
Sens. 2020, 12, x FOR PEER REVIEW 4 of 20
If the elevation of the area of interest changes significantly, flying at constant height ATO does
not provide
If the optimal
elevation image
of theoverlap
area offor 3D reconstruction.
interest However,flying
changes significantly, we see
atthe following
constant solutions
height to
ATO does
the problem:
not provide optimal image overlap for 3D reconstruction. However, we see the following solutions
to the problem:
• A combination of several missions flown at different altitudes,
• A combination of several missions flown at different altitudes,
• Fly the UAV manually or,
• Fly the UAV manually or,
• Use a flight planning application that allows terrain following.
• Use a flight planning application that allows terrain following.
The
The first
first option,
option, a combination
a combination of several
of several missions,
missions, is moreistime-consuming
more time-consuming thanmission,
than a single a single
and you must be careful to prevent gaps between the areas flown. Manual flying is even moreistime
mission, and you must be careful to prevent gaps between the areas flown. Manual flying even
consuming, it is difficult to maintain the same height AGL, and there is no guarantee of correct imageof
more time consuming, it is difficult to maintain the same height AGL, and there is no guarantee
correct Ifimage
overlap. overlap.
an adequate If anmodel
terrain adequate terrain an
is available, model is available,
autonomous flightan autonomous
mission following flight mission
the terrain
following the
is the best option. terrain is the best option.
AAdigital
digitalterrain
terrain model
model (DTM)(DTM) isisrequired
requiredtoto enable
enablethethe
UAV UAV to follow the non-flat
to follow terrain.
the non-flat DTMs
terrain.
or DEMs with a certain spatial resolution can be provided worldwide (mostly
DTMs or DEMs with a certain spatial resolution can be provided worldwide (mostly in low resolution), in low resolution),
nationwide
nationwide (mostly
(mostly in in different
different resolutions)
resolutions) andand locally
locally fromfrom previous
previous surveys.
surveys. TheThe biggest
biggest problem
problem is
is lack
the the lack of software
of software that supports
that supports terrainterrain
data fordata for mission
mission planning.
planning. At the
At the time timesurvey
of our of ourflights
survey
flights
(end (endwe
of 2018), of found
2018), awe found
single a single for
application application
DJI UAVs for withDJI
DTMUAVs with
import DTM import
capabilities, capabilities,
namely Drone
namely Drone Harmony. According to [7] from 2019, UgCS
Harmony. According to [7] from 2019, UgCS was the only other commercial application that was the only other commercial
offered
application
such thatat
possibilities offered such possibilities at that time.
that time.

2.3.Basic
2.3. Basicaspects
aspectsofofUAV
UAV Photogrammetry
Photogrammetry inin Steep
Steep Terrain
Terrain
AnAn important
important aspect
aspect of of
UAVUAV photogrammetry
photogrammetry in steep
in steep terrain
terrain is the
is the direction
direction of theof optical
the optical
axis axis
of
of the camera. The image scale in aerial surveying is based on the ratio between
the camera. The image scale in aerial surveying is based on the ratio between the distance from the the distance from the
image
image element
element toto
thethe projection
projection centre
centre andandthethe distance
distance from
from thethe projection
projection centre
centre totothethe object
object onon
thethe
terrain.When
terrain. Whenthe theoptical
opticalaxis
axisofof the
the camera
camera waswas directed
directed vertically
vertically toto
thethe ground,
ground, as asis is common
common inin
flat
flat areas,
areas, the
the image
image scale
scale ofof steep
steep terrain
terrain changed
changed significantly.InInthese
significantly. thesecases,
cases,the theoblique
obliqueimages,
images,
whose
whose optical
optical axes
axes were
were perpendicular
perpendicular toto
thethe ground
ground surface,
surface, provided
provided much
much better
better image
image properties
properties
withrespect
with respecttotothe
theimage
imagescale
scaleandandthethe subsequentimage
subsequent imageGSD.GSD.Figure
Figure2 2displays
displaysthe thegeometric
geometric
comparison
comparison between
between the
the vertical
vertical image
image andandthethe oblique
oblique image
image perpendicular
perpendicular to to
thethe terrain.
terrain.

(a) (b)

Figure Photographing
2. 2.
Figure steep
Photographing terrain
steep with
terrain a (a)
with vertical
a (a) image
vertical and
image anan
and (b)(b)
oblique image.
oblique image.

An example of an extreme situation on a vertical image in steep terrain is shown in Figure 3. The
height differences from the image projection centre to points A and B were 288.4 and 19.6 meters,
respectively. The 3D distances to the same points were 362.0 and 53.7 meters.
 Remote Sens. 2020, 12, 1293 5 of 20

An example of an extreme situation on a vertical image in steep terrain is shown in Figure 3.

The height differences from the image projection centre to points A and B were 288.4 and 19.6 meters,
respectively. The 3D distances to the same points were 362.0 and 53.7 meters.
Remote Sens. 2020, 12, x FOR PEER REVIEW 5 of 20

Figure 3. Extreme height differences in a steep terrain result in a varying GSD within the vertical
Figure 3. Extreme height differences in a steep terrain result in a varying GSD within the vertical image
image (~0.5 cm in point A and ~7.9 cm in point B).
(~0.5 cm in point A and ~7.9 cm in point B).
Since the image overlap is calculated according to the planned UAV height AGL, there is an
Since the image overlap is calculated according to the planned UAV height AGL, there is an issue
issue in vertical images because the distance to the upper part of the captured area in the image is
in vertical images because the distance to the upper part of the captured area in the image is much
much shorter than the height AGL. In Figure 3, point B was only 19.6 meters below and 53.7 meters
shorter than the height AGL. In Figure 3, point B was only 19.6 meters below and 53.7 meters away
away from the UAV, whose projected height above the DTM was 80 meters. As a result, the overlap
from the UAV, whose projected height above the DTM was 80 meters. As a result, the overlap was
was reduced and could become critically low in very steep areas. The situation is depicted in Figure
reduced and could become critically low in very steep areas. The situation is depicted in Figure 4 for
4 for vertical images and horizontal flat terrain. Equation (1) provides the relationships between the
vertical images and horizontal flat terrain. Equation (1) provides the relationships between the overlap
overlap fraction (R), the image base (B) and the dimensions of the image sensor scaled to the ground
fraction (R), the image base (B) and the dimensions of the image sensor scaled to the ground (W) [23]:
(W) [23]:
𝐵B
𝑅 R==1 1−− W (1) (1)
𝑊
TheThe image
image basebase was calculated
was calculated from
from the the specified
specified height
height AGL andAGL and the overlap.
the projected projectedWith
overlap.
a
With aimage
constant constant
base,image base, the
the overlap wasoverlap
reducedwastoreduced to 60%
60% at the half at
of the
the half of the
height AGL height AGL and
and further
further
reduced toreduced
20% for to 20% forofa the
a quarter quarter of the
height AGL.height AGL.
(W) [23]:
𝐵
𝑅 = 1 − (1)
𝑊
The image base was calculated from the specified height AGL and the projected overlap. With a
constant
Remote Sens.image
2020, 12,base,
the overlap was reduced to 60% at the half of the height AGL and further
1293 6 of 20
reduced to 20% for a quarter of the height AGL.

Remote Sens. 2020, 12, x FOR PEER REVIEW 6 of 20

Figure 4. Decreased overlap in steep terrain.
Figure 4. Decreased overlap in steep terrain.
2.4. Development of Custom DEMs
2.4. Development of Custom DEMs
The creation of custom DEMs is applicable to any publicly available DTM. In our experience,
Theagainst
we advise creationusing
of custom DEMs isSRTM
the global applicable to any
(shuttle publicly
radar available mission)
topography DTM. In our experience,
DEM. Due to weits spatial
advise against using the global SRTM (shuttle radar topography mission) DEM. Due to its spatial
resolution of 3 arcseconds, the SRTM DEM is not accurate enough to plan a UAV mission in steep hilly
resolution of 3 arcseconds, the SRTM DEM is not accurate enough to plan a UAV mission in steep
terrain [9,24–26]. The national 5-meter DTM (NDTM) of The Surveying and Mapping Authority of
hilly terrain [9,24–26]. The national 5-meter DTM (NDTM) of The Surveying and Mapping Authority
Slovenia was was
of Slovenia selected for for
selected thethe
study.
study.

2.4.1. The
2.4.1. Plane
The PlaneDEM
DEM
ToTo
prevent
preventthe UAV
the UAVfromfromgetting
gettingtoo
too close
close to an area
to an areathat
thatsuddenly
suddenlybecame
became very
very steep—since the
steep—since
UAVtheheight AGL AGL
UAV height is theisvertical distance
the vertical from
distance thethe
from imported
importedNDTM,
NDTM,see see Figure 5a—theidea
Figure 5a—the ideawas
was born
born toacreate
to create planeasurface
plane surface overextreme
over the the extreme parts
parts of the
of the area.
area. Consequently, the
Consequently, the distance
distanceofofthe
the UAV
UAV from the terrain was much more favorable, as shown
from the terrain was much more favorable, as shown in Figure 5b. in Figure 5b.

(a) (b)

Figure 5. 5.Oblique
Figure Obliqueimages acquiredwhen
images acquired whenthethe angle
angle of inclination
of inclination changes
changes suddenly:
suddenly: (a) use (a) use of the
of the
national 5-meter
national 5-meterdigital
digital terrain model(NDTM)
terrain model (NDTM) forfor mission
mission planning
planning anduse
and (b) (b)ofuse
the of the constructed
constructed
plane
plane digitalelevation
digital elevationmodel
model (DEM)
(DEM)for
formission
missionplanning.
planning.

Figure 6 displays the plane DEM set above the NDTM. Three reference points were selected for
the calculation of the plane. Two reference points, marked 1 and 2 in Figure 6, were located in the
lowest part of the NDTM, and point 3 was the one where the plane through points 1 and 2 touched
the NDTM as it approached the NDTM from the vertical position. The reference XY coordinates of
 Remote Sens. 2020, 12, 1293 7 of 20

Figure 6 displays the plane DEM set above the NDTM. Three reference points were selected for
the calculation of the plane. Two reference points, marked 1 and 2 in Figure 6, were located in the
lowest part of the NDTM, and point 3 was the one where the plane through points 1 and 2 touched
the NDTM as it approached the NDTM from the vertical position. The reference XY coordinates of
the plane DEM were the same as for the NDTM, only the elevations were recalculated. Therefore the
resolution
Remote of12,
Sens. 2020, the plane
x FOR DEM
PEER was the same as for the NDTM.
REVIEW 7 of 20

Figure 6. The plane DEM (in red) of the case study above the NDTM (in green) with the reference points.
Figure 6. The plane DEM (in red) of the case study above the NDTM (in green) with the reference
points.
This DEM ensures a certain minimum distance to the terrain. Furthermore, an appropriate
overlap was ensured. The disadvantage is the greater distance to some areas, e.g., local depressions,
Thisleads
which DEMtoensures
a lower amodel
certain minimum
resolution distance to the terrain. Furthermore, an appropriate
of them.
overlap was ensured. The disadvantage is the greater distance to some areas, e.g. local depressions,
2.4.2.leads
which The to
Fake DEMmodel resolution of them.
a lower
The plane DEM is a simple surface that is easy to calculate because only three reference points are
2.4.2. The Fake DEM
needed. However, this simplicity has some shortcomings. The differences in elevations between the
TheDEM
plane planeandDEM theisDTM
a simple surface large.
can become that isTherefore
easy to calculate
anotherbecause
custom only
DEMthree reference points
was developed.
are needed. However,
The fake DEM isthis simplicity
specially has some
designed shortcomings.
for UAV surveying The
withdifferences in elevations
oblique images. The fakebetween
DEM grid
theisplane DEM and the DTM can become large. Therefore another custom DEM was developed.
shifted to the NDTM grid so that the DCG to the NDTM is constant at given values of the projected
The fake
height above DEMthe is specially
fake DEM, UAVdesigned for UAV
heading surveying
and gimbal with
angle, alloblique images.
of which The fake
are constant DEMagrid
during flying
is shifted
mission.toThe
the NDTM grid so
recalculated gridthat the DCG
points to the
and their NDTM isform
elevations constant at given
the fake DEM.values of the projected
A graphical relationship
height above
between thethe fake
fake DEM,
DEM UAV
and the heading
NDTM isand gimbal
shown angle, all
in Figure 7. of which are constant during a flying
mission. The recalculated grid points and their elevations form the fake DEM. A graphical
2.4.3. The between
relationship tilted DEMs the fake DEM and the NDTM is shown in Figure 7.
An unexpected UAV's behavior was observed during its practical deployment in a mountainous
terrain. As the UAV height ATO increased, the height AGL decreased. A generalized situation is
shown in Figure 8a. The UAV starts the survey at the projected height AGL h_0. The height AGL at
the top, denoted h_top, should be similar to h_0. However, in our case study, which is described in
Section 3, it was not, since h_top was in all cases about 20 meters smaller than h_0. To compensate for
the error, the tilted DEMs were introduced. The idea is to tilt a DEM so that the actual flight height
is more parallel to the basic DEM and h_top is similar to h_0. The situation is shown in Figure 8b.
The compensation value (CV) has been set at 30 meters, to be on the safe side. The tilted DEMs were
recalculated DEMs so that the elevation increased linearly from 0 at the bottom to 30 meters at the top.

(a) (b)
 2.4.2. The Fake DEM
The plane DEM is a simple surface that is easy to calculate because only three reference points
are needed. However, this simplicity has some shortcomings. The differences in elevations between
the plane DEM and the DTM can become large. Therefore another custom DEM was developed.
Remote The
Sens. fake DEM
2020, 12, 1293is specially designed for UAV surveying with oblique images. The fake DEM grid
8 of 20
is shifted to the NDTM grid so that the DCG to the NDTM is constant at given values of the projected
height above the fake DEM, UAV heading and gimbal angle, all of which are constant during a flying
The tilted values
mission. were applied
The recalculated to the
grid NDTM,
points andthe plane
their DEM andform
elevations the fake
the DEM. Figure A
fake DEM. 8c graphical
displays
an example of a 3D comparison of NDTM and the tilted NDTM.
relationship between the fake DEM and the NDTM is shown in Figure 7.

Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 20

planning. When approaching a steep feature, the fake DEM guides UAV to back off from the terrain
to keep the DCG constant.

2.4.3. The tilted DEMs

An unexpected UAV's behavior was observed during its practical deployment in a mountainous
terrain. As the UAV height ATO increased, the height AGL decreased. A generalized situation is
shown in Figure 8a. The UAV starts the survey at the projected height AGL h_0. The height AGL at
the top, denoted h_top, should be similar to h_0. However, in our case study, which is described in
Section 3, it was not, since h_top was in all cases about 20 meters smaller than h_0. To compensate
for the error, the tilted DEMs were introduced. The idea is to tilt a DEM so that the actual flight height
is more parallel to the basic DEM and h_top is similar to h_0. The situation is shown in Figure 8b.
The compensation value (CV) has (a)been set at 30 meters, to be on the safe side. The(b) tilted DEMs were
recalculated DEMs so that the elevation increased linearly from 0 at the bottom to 30 meters at the
Figure
top. The
Figure 7.7.Fake
tilted Fake DEMwere
values
DEM andNDTM:
and NDTM:
applied(a)to
(a) Graphical
the NDTM,
Graphical 3Dcomparison
3D comparison
the plane offake
of fakeDEM
DEM DEM (blue)
and(blue) andNDTM
the fake
and NDTM (green)
DEM.(green)
Figure 8c
ofofthe
displays the case
ancase study.The
example
study. The
of fakecomparison
a fake
3D DEMisisshifted
DEM shifted soatat80
ofso
NDTM 80m
m above
and the
thethe
above fakeNDTM.
tilted
fake DEMthe
DEM thedistance
distancecamera-ground
camera-ground
(DCG)atatpitch
(DCG) pitchangle—45
angle—45degrees
degreesisis60
60mmand
and(b)
(b)oblique
obliqueimage
imageacquired
acquiredusing
usingfake
fakeDEM
DEMininmission
mission
planning. When approaching a steep feature, the fake DEM guides UAV to back off from the terrain to
keep the DCG constant.

(a)
(b)

Figure 8. Cont.
Remote Sens.
Remote 2020,
Sens. 12,12,
2020, 1293
x FOR PEER REVIEW 9 of920
of 20

(c)
Figure
Figure 8. 8.DEMs
DEMsand
andtilted
tiltedDEMs:
DEMs: (a)
(a) UAV
UAV flight
flight heights
heightswhen
whenthe
thebasic
basicDEM
DEMis isused
usedforfor
mission
mission
planning; (b) UAV flight heights when the tilted DEM is used for mission planning
planning; (b) UAV flight heights when the tilted DEM is used for mission planning and (c) and (c) a 3D
a 3D
comparison of NDTM (green) and tilted NDTM (dark orange) of the case study.
comparison of NDTM (green) and tilted NDTM (dark orange) of the case study.

3. 3. Site
Site Description,Field
Description, FieldWork
Workand
andData
Data Processing
Processing
The
The developed
developed DEMs
DEMs (plane,
(plane, fake,fake,
tiltedtilted plane,
plane, tiltedtilted faketilted
fake and and NDTM)
tilted NDTM) and terrain-
and terrain-following
following aerial surveys were tested on the Belca active rockfall in the NW
aerial surveys were tested on the Belca active rockfall in the NW part of Slovenia. It lies onlypart of Slovenia. It lies
a few
only a few 100 m above the village of Belca, while a sawmill is located just below the
100 m above the village of Belca, while a sawmill is located just below the unstable slopes. In 1953 unstable slopes.
In 1953
a large a largeevent
rockfall rockfall
withevent with considerable
considerable damage damage
was reportedwas reported in thisAarea.
in this area. forestA road
forestcrossing
road
crossing the area is shown in Figure 9a. The next major rockfall event in February 2018 triggered the
the area is shown in Figure 9a. The next major rockfall event in February 2018 triggered the debris
debris flow that accumulated on the road that has been closed since then. Another massive rockfall
flow that accumulated on the road that has been closed since then. Another massive rockfall event
event occurred in October 2018, eroding the left bank of the Belca torrent in addition to the material
occurred in October 2018, eroding the left bank of the Belca torrent in addition to the material on the
on the rockfall area. The debris has accumulated in and around the Belca river bed. It has largely
rockfall area. The debris has accumulated in and around the Belca river bed. It has largely damaged
damaged the smaller hydroelectric power station and the local sawmill. Thereafter, some cracks in
thethe
smaller hydroelectric power station and the local sawmill. Thereafter, some cracks in the rock
rock mass near the top of the slide caused concern, so that a controlled removal of the threatening
mass near the top of thewas
block with explosives slide caused The
initiated. concern, so extends
rockfall that a controlled
over an arearemoval of the
of about 100 threatening
m × 300 m and block
with explosives was initiated. The rockfall extends over an area
spreads between elevations 710 m AMSL at the toe and 1020 m AMSL at the crown.of about 100 m × 300 m and spreads
between elevations 710 m AMSL at the toe and 1020 m AMSL at the crown.
Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 20

(a) (b)
Figure 9. Belca rockfall: (a) on the national orthophoto and (b) on a photograph taken at the bottom.
Figure 9. Belca rockfall: (a) on the national orthophoto and (b) on a photograph taken at the bottom.
In the period from November to December 2018, several UAV surveys were carried out in which
different DEMs and types of images were tested. Table 1 shows an overview of the surveys.

Table 1. Surveys of the Belca rockfall with information on DEMs and the type of images used in each
mission.
Remote Sens. 2020, 12, 1293 10 of 20

In the period from November to December 2018, several UAV surveys were carried out in which
different DEMs and types of images were tested. Table 1 shows an overview of the surveys.

Table 1. Surveys of the Belca rockfall with information on DEMs and the type of images used in
each mission.

Date DEM/Imagery DEM/Imagery DEM/Imagery

8 Nov 2018 NDTM/oblique / /
4 Dec 2018 plane/oblique plane/oblique fake/oblique
11 Dec 2018 tilted_NDTM/vertical tilted_plane/oblique tilted_fake/oblique

The UAV height AGL dictates the GSD of the images. The DJI Phantom 4 Pro used in research
achieved a GSD of 1 cm at a height of 36.5 m, 2 cm at a height of 73 m, etc.
The NDTM was geolocated in the plane coordinate system in XYZ format. For use with the
Drone Harmony app, it must be transformed into WGS-84 and then exported in Esri ASCII format.
The NDTM was initially used for vertical and oblique image acquisition. For the latter, the gimbal
pitch was set to -45 degrees, as the average slope angle is close to 45 degrees. The planned height
AGL for all flights was set at 80 meters. The front overlap was set to 80% and a side overlap at 65%.
As already mentioned in Section 2.3, the image overlap is reduced in steep terrain. To keep the actual
overlap at the desired rate, the projected overlap can be set higher, but this would mean more images
and longer flight times. In order to facilitate the comparison of the two sets of images (vertical and
oblique), it was decided to keep the same projected heights and overlaps.
Since the topography of the area is diverse and the slope angle changes rapidly, the NDTM was
considered unsuitable for oblique imagery. Two user-defined DEMs were created to support the
acquisition of the oblique images.
The plane DEM was generated in the national coordinate system and transformed into WGS84
and Esri ASCII format, same as for the NDTM. The steepest inclination of the plane DEM was 43.9
degrees with an azimuth of 332 degrees. Since the surface of the plane DEM differs from the NDTM,
the GSD on the NDTM varies when the plane DEM was used in mission planning.
For the fake DEM of the Belca case, the elevations of the DTM were recalculated in such a way
that from any point 80 meters above the DEM, the DCG at the pitch angle -45 degrees and heading
330 degrees was 60 meters. In practical use, the fake DEM was uploaded to the mission control app,
the flight height AGL was set to 80 meters, the gimbal pitch angle to -45 degrees and the heading to
330 degrees. The DCG to the NDTM terrain was 60 meters. The value of 60 meters was a rounded
value of 80 m × cos (45◦ ) = 56.6 m, where 45◦ was the pitch angle.
The georeferencing of the UAV images was performed using ground control points (GCPs),
which were signaled with targets (a black circle 27 cm in diameter on a white background). Ideally,
the GCPs should be evenly distributed over a surveyed area [27–29] to achieve optimal accuracy of
results. The terrain of the Belca rockfall is not only steep, but it is also a challenge to reach some areas in
order to set the GCPs optimally. Figure 10 shows the geometric distribution of the GCPs. In the middle
of the area, in this case in the central part of the rockfall, there should be some GCPs. Firstly, it is very
dangerous to cross the rocky area and secondly, it would be difficult to fix a GCP target on the unstable
ground. Since some of the targets were set under adverse GNSS conditions, we used two different
GNSS receivers and two modes, RTK and fast-static, to achieve high quality results. The acquired
accuracy of the GCP positions was about 5 cm. The GCP coordinates were determined in the national
reference coordinate system D96/TM (EPSG: 3794).
middle of the area, in this case in the central part of the rockfall, there should be some GCPs. Firstly,
it is very dangerous to cross the rocky area and secondly, it would be difficult to fix a GCP target on
the unstable ground. Since some of the targets were set under adverse GNSS conditions, we used two
different GNSS receivers and two modes, RTK and fast-static, to achieve high quality results. The
Remote Sens.accuracy
acquired 2020, 12, 1293
of the GCP positions was about 5 cm. The GCP coordinates were determined 11 ofin
20
the national reference coordinate system D96/TM (EPSG: 3794).

Figure
Figure10.
10.Distribution
Distributionof
ofGCPs
GCPs(shown
(shownon
onthe
thedense
densepoint
pointcloud).
cloud).

The imageswere
The images were processed
processed withwith Agisoft
Agisoft Metashape
Metashape softwaresoftware
[30]. The[30]. The
images' images'
exterior exterior
orientations,
orientations, calculated
calculated during duringblock
the bundle the bundle block (BBA)
adjustment adjustment
were (BBA)
used towere usedthe
analyze to analyze theofdistances
distances the UAV
of thethe
from UAV from Together
ground. the ground.
withTogether
the BBA, with
densethe BBA,
point dense
clouds andpoint clouds and
orthophotos wereorthophotos were
created to visually
created
evaluatetothe
visually
effectsevaluate theflight
of various effects of various
missions flight
on the missions
final on the final products.
products.

4.Results
4. Results
Thissection
This sectionpresents
presentsthe
themain
mainheight-related
height-relatedresults
resultsof
ofthe
theBelca
Belcacase
casestudy,
study, which
whichused
usedcustom
custom
DEMsthat
DEMs thatwould
wouldensure
ensureaaconstant
constantdistance
distancebetween
betweenthe
theUAV
UAV and
and the
the terrain
terrain in
inaavertical
verticalor
oroblique
oblique
direction. The
direction. Thepractical
practicalaspects
aspectsof
ofvertical
verticaland
andoblique
obliqueimages
images were
were discussed
discussed in in Section
Section 5.2.
5.2.

4.1.UAV
4.1. UAVheights
heightsover
overNDTM
NDTMand
andcustom
customDEMs
DEMs
DJI's technical
DJI's technical support
supportwould
wouldnot notreveal
reveal to
to us
us the
the principles
principles behind
behind the
the estimation
estimation ofof aircraft
aircraft
height. Based
height. Based onon information
informationcollected
collectedinin
thethe
forum on the
forum on DJI
the website and [31],
DJI website andDJI's
[31],non-RTK aircraft
DJI's non-RTK
operate with relative heights, using the on-board barometric altimeter and height 0 being the point at
which the UAV takes off from the ground.
When using the terrain file in Drone Harmony Pro application, it is necessary to select the starting
point, as the height at this point is set to 0. According to [31] the mission planning procedure creates
a grid over the user selected polygon. Each grid point height was calculated as a weighted average
over the adjacent DTM grid points.
In Figures 11–15, the UAV height ATO refers to the barometric height of the UAV at the time
an image is taken. The height AGL was calculated as the height ATO, reduced by the interpolated
elevation value of a DEM at the point vertically below the projection centre (PC) of an image.
The barometric height ATO was taken from the flight log. The coordinates of PCs for all images were
 starting point, as the height at this point is set to 0. According to [31] the mission planning procedure
creates a grid over the user selected polygon. Each grid point height was calculated as a weighted
average over the adjacent DTM grid points.
In Figures 11-15, the UAV height ATO refers to the barometric height of the UAV at the time an
image is taken. The height AGL was calculated as the height ATO, reduced by the interpolated
Remote Sens. 2020, 12, 1293 12 of 20
elevation value of a DEM at the point vertically below the projection centre (PC) of an image. The
barometric height ATO was taken from the flight log. The coordinates of PCs for all images were
estimated
estimated in in the
the bundle
bundle adjustment
adjustment during
during thethe SfM.
SfM. The
The DCGs
DCGs toto DEMs
DEMs were
wereanalyzed
analyzed either
either ininthe
the
vertical direction or in the slant direction of the optical
vertical direction or in the slant direction of the optical axis. axis.
Figure
Figure 11 11 shows
shows thethe UAV
UAVheights
heightsabove
above the
theNDTM
NDTM using
using NDTM
NDTM or or custom
custom DEM
DEM inin mission
mission
planning.
planning.When Whenusing usingNDTM,
NDTM, some of the
some of spikes couldcould
the spikes be caused by NDTM
be caused smoothing
by NDTM in the mission
smoothing in the
planning application. Since the surface of the plane DEM is smooth by itself, sudden changes
mission planning application. Since the surface of the plane DEM is smooth by itself, sudden changes in height
are smallerare
in height than whenthan
smaller using the NDTM
when using theforNDTM
missionfor
planning.
mission planning.

Figure 11. UAV

Figure11. UAVflight
flightheights
heightsabove
abovethetheNDTM
NDTMusing
usingNDTM
NDTMin
inmission
missionplanning
planning(blue)
(blue) and
andusing
using
plane DEM in mission planning (red).
plane DEM in mission planning (red).

The flight heights of the same plane DEM based flight over the plane DEM are shown in Figure 12a.
The flight heights of the same plane DEM based flight over the plane DEM are shown in Figure
The corresponding image block is shown in Figure 12b. The first spike (around image number 140) was
12a. The corresponding image block is shown in Figure 12b. The first spike (around image number
a consequence of the PC heights of the images named 297-300, which should be almost equal, but the
140) was a consequence of the PC heights of the images named 297-300, which should be almost
UAV descended 12 meters in between. The last spike was located between the images named 310 and
equal, but the UAV descended 12 meters in between. The last spike was located between the images
Remote
311.Sens.
The2020, 12, x FOR PEER
difference REVIEW
in height should be more than 15 meters, but the actual lift was only 4.513meters.
of 20
named 310 and 311. The difference in height should be more than 15 meters, but the actual lift was
only 4.5 meters.

(a) (b)
Figure 12. (a)
Figure 12. UAV flight
(a) UAV height
flight above
height the the
above plane DEM
plane andand
DEM (b) (b)
horizontal image-block.
horizontal Labels
image-block. are are
Labels
image filenames.
image filenames.

TheThe height
height of the
of the UAV UAV above
above a certain
a certain DEM DEM should
should remain
remain constant
constant throughout
throughout the mission.
the mission.
Ideally,
Ideally, thisthisheight
heightshould
should bebe8080meters, as itaswas
meters, it planned for all flights.
was planned for all Due to unknown
flights. Due toinconsistencies
unknown
of the initialof
inconsistencies height ATO the
the initial starting
height ATOheight deviated
the starting fromdeviated
height 80 meters. fromHowever, the following
80 meters. However, heights
the
AGL should be close to the initial height. The cases presented differ from the
following heights AGL should be close to the initial height. The cases presented differ from the expected ones because
expected ones because the height AGL decreased steadily with increasing amount of the height ATO.
All other flights of the first two days show the same trend.
Even though the DCGs are only relevant for the actual terrain represented by the NDTM, the
example in Figure 12 is presented to further strengthen the hypothesis that the barometric height
AGL of the UAV differed from the actual height AGL.
To investigate the issue, the PCs' heights ATO of one flight were compared with the barometric
 Figure 12. (a) UAV flight height above the plane DEM and (b) horizontal image-block. Labels are
image filenames.

The height of the UAV above a certain DEM should remain constant throughout the mission.
Remote Sens.this
Ideally, 12, 1293
2020,heightshould be 80 meters, as it was planned for all flights. Due to unknown 13 of 20
inconsistencies of the initial height ATO the starting height deviated from 80 meters. However, the
following heights AGL should be close to the initial height. The cases presented differ from the
the height AGL decreased steadily with increasing amount of the height ATO. All other flights of the
expected ones because the height AGL decreased steadily with increasing amount of the height ATO.
first two days show the same trend.
All other flights of the first two days show the same trend.
Even though the DCGs are only relevant for the actual terrain represented by the NDTM,
Even though the DCGs are only relevant for the actual terrain represented by the NDTM, the
the example in Figure 12 is presented to further strengthen the hypothesis that the barometric height
example in Figure 12 is presented to further strengthen the hypothesis that the barometric height
AGL of the UAV differed from the actual height AGL.
AGL of the UAV differed from the actual height AGL.
To investigate the issue, the PCs' heights ATO of one flight were compared with the barometric
To investigate the issue, the PCs' heights ATO of one flight were compared with the barometric
heights ATO of the UAV of the same flight. The adjusted PC positions can be considered as the true
heights ATO of the UAV of the same flight. The adjusted PC positions can be considered as the true
positions. The comparison is depicted in Figure 13. The deviation increased to about 20 meters at the
positions. The comparison is depicted in Figure 13. The deviation increased to about 20 meters at the
largest heights ATO, which was consistent with the cases shown above.
largest heights ATO, which was consistent with the cases shown above.

Figure13.
Figure 13.Differences
Differencesofof
thethe barometric
barometric heights
heights andand the actual
the actual heights
heights aboveabove take-off
take-off (ATO).(ATO). The
The values
values
in in theare
the figure figure are calculated
calculated as the barometric
as the barometric heights heights subtracted
subtracted by the
by the PC's PC's ATO.
heights heights ATO.

The flights
The flights were
were carried
carried out
out under
under different
differentweather
weatherconditions.
conditions. Most
Most of
of the
the flights
flights were
were from
from
bottom to top, some also from top to bottom. The height difference at the top was always in
bottom to top, some also from top to bottom. The height difference at the top was always in the range the range
of 15-25
of 15-25 meters,
meters,sosothat
thatititcan
canbe
beconsidered
consideredaasystematic
systematicerror.
error.

4.2.
4.2. Tilted
Tilted DEMs
DEMs
To solve the issue, the tilted DEMs were developed, see Section 2.4. As an example, the heights
above the fake DEM and the tilted fake DEM are shown in Figure 14. The actual height above the DEM
decreased again as the height ATO increased. However, the heights above the original fake DEM were
much more constant. For comparison, the barometric heights of the UAV are shown.
Remote Sens. 2020, 12, x FOR PEER REVIEW 14 of 20
Remote Sens. 2020, 12, x FOR PEER REVIEW 14 of 20

To solve the issue, the tilted DEMs were developed, see Section 2.4. As an example, the heights
To solve the issue, the tilted DEMs were developed, see Section 2.4. As an example, the heights
above the fake DEM and the tilted fake DEM are shown in Figure 14. The actual height above the
above
Remote
the2020,
fake12,DEM and the tilted fake DEM are shown in Figure 14. The actual height above the
DEM Sens.
decreased 1293 as the height ATO increased. However, the heights above the original
again 14 fake
of 20
DEM decreased again as the height ATO increased. However, the heights above the original fake
DEM were much more constant. For comparison, the barometric heights of the UAV are shown.
DEM were much more constant. For comparison, the barometric heights of the UAV are shown.

Figure 14. Heights

Heights above
above the fake DEM
DEM (blue), heights above the tilted fake DEM (red) and barometric
Figure
Figure 14.
14. Heights above the
the fake
fake DEM (blue),
(blue), heights
heights above
above the
the tilted
tilted fake
fake DEM
DEM (red)
(red) and
and barometric
barometric
heights
heights (orange)
(orange) above
above the
the fake
fake DEM.
DEM.
heights (orange) above the fake DEM.

If the tilted DEMs were used for mission planning, the actual heights AGL were much more
If the tilted DEMs were used for mission planning, the actual heights AGL were much more
constant, as can be clearly seen in Figure 14 (see
(see blue
blue line).
line).
constant, as can be clearly seen in Figure 14 (see blue line).
Moreover,
Moreover, whenwhenusing
usingoblique
obliqueimages,
images,thethe DCGs
DCGs along
along thethe optical
optical axisaxis
werewere
moremore important
important than
Moreover, when using oblique images, the DCGs along the optical axis were more important
than the actual flight heights AGL. The DCGs of all three missions on the third day to the
the actual flight heights AGL. The DCGs of all three missions on the third day to the actual terrain actual
than the actual flight heights AGL. The DCGs of all three missions on the third day to the actual
terrain represented
represented by the NDTM
by the NDTM are in
are shown shown
Figurein15.
Figure 15.
terrain represented by the NDTM are shown in Figure 15.

Figure 15. Slant DCGs to the NDTM from the tilted fake DEM (green), tilted plane DEM (blue) and
Figure 15.
Figure 15. Slant
Slant DCGs
DCGs toto the
the NDTM
NDTM from
from the
the tilted
tilted fake
fake DEM
DEM (green),
(green), tilted
tilted plane
plane DEM
DEM (blue)
(blue) and
and
tilted NDTM (red). The slant distances from the tilted NDTM are simulated with the values of pitch
tilted NDTM
tilted NDTM (red).
(red). The slant distances from the
the tilted
tilted NDTM
NDTM are
are simulated
simulated withwith the
the values
values of
of pitch
pitch
angle – 45 degrees at azimuth 330 degrees.
angle –– 45
angle 45 degrees
degrees at
atazimuth
azimuth330330degrees.
degrees.

The DCGs from the tilted fake DEM mission are the most constant, ranging from 47 to 72 meters.
The
The DCGs
DCGs from
from the
the tilted
tilted fake
fake DEM mission are the most constant, ranging from 47 to 72 meters.
As a reminder, the fake DEM was designed so that the DCGs are at least theoretically 60 meters long.
As
As aa reminder,
reminder,the
thefake
fakeDEM
DEM waswas designed
designed so
so that
that the
the DCGs
DCGs are
are at
at least theoretically
theoretically 60 meters long.
5.
5. Discussion
5. Discussion
Discussion
5.1. UAV over-Terrain Flying in Steep Terrain
Terrain
5.1. UAV over-Terrain Flying in Steep Terrain
To execute a terrain-following mission survey, a special software application and a DEM are
required. When flying in steep terrain, the flight planning software must allow the user to import
Remote Sens. 2020, 12, x FOR PEER REVIEW 15 of 20
Remote Sens. 2020, 12, 1293 15 of 20
To execute a terrain-following mission survey, a special software application and a DEM are
required. When flying in steep terrain, the flight planning software must allow the user to import an
an accurate terrain description [7]. We used Drone Harmony application as one of the few that were
accurate terrain description [7]. We used Drone Harmony application as one of the few that were able
able to import a DEM for a DJI-made UAV at the time of our surveys.
to import a DEM for a DJI-made UAV at the time of our surveys.
Figure 16 shows the comparison between the SRTM DEM and the NDTM. Since the SRTM DEM
Figure 16 shows the comparison between the SRTM DEM and the NDTM. Since the SRTM DEM
generalizes the terrain due to the low resolution, it is clear that the SRTM DEM cannot be used for
generalizes the terrain due to the low resolution, it is clear that the SRTM DEM cannot be used for
UAV missions in steep hilly terrain. It could even become dangerous if the aircraft follows the SRTM
UAV missions in steep hilly terrain. It could even become dangerous if the aircraft follows the SRTM
terrain, because it could crash to the ground. The SRTM DEM was more than 75 m below the ground
terrain, because it could crash to the ground. The SRTM DEM was more than 75 m below the ground
at some spots of our test area, see Figure 16b. The authors of [9] also claim that the global DTMs may
at some spots of our test area, see Figure 16b. The authors of [9] also claim that the global DTMs may
not be able to provide sufficient accuracy in the estimation of height AGL.
not be able to provide sufficient accuracy in the estimation of height AGL.

(a) (b)
Figure
Figure16.
16.Comparison
Comparisonofof
SRTM
SRTMDEMDEMand andNDTM:
NDTM:(a)(a)Spatial
Spatialcomparison
comparisonofofSRTM
SRTMDEMDEMiningreen
green
tones
tonesand
andNDTM
NDTMininbrown
browntones
tonesand
and(b)
(b)hypsometric
hypsometricdisplay
displayofofheight
heightdifferences,
differences,which
whichrange
rangefrom
from -
- 75
75 to
to 100
100 meters.

5.2.Vertical
5.2. Verticaland
andOblique
ObliqueImages
ImagesininSteep
SteepTerrain
Terrain
Theuse
The useofofoblique
oblique images
images isis particularly
particularly appropriate
appropriateininhillyhillyterrain
terrainwith
withrugged
ruggedtopography
topography and
overhangs, as shown in Figure 17. The images in Figure 17a–c show dense point
and overhangs, as shown in Figure 17. The images in Figure 17a, 17b and 17c show dense point clouds clouds from single
surveys,
from singlewhile Figure
surveys, 17dFigure
while displays 17da displays
merged point
a mergedcloudpoint
fromcloud
the vertical
from theimages andimages
vertical the oblique
and
images
the taken
oblique on the
images plane
taken onDEM. If only
the plane the vertical
DEM. If only images are used,
the vertical images many
aretopographic
used, many details may be
topographic
lost, asmay
details shown in Figure
be lost, 17a. in
as shown With the 17a.
Figure use of the the
With oblique
use ofimages and the
the oblique combination
images and the of vertical and
combination
of vertical and oblique images, see Figure 17d, the result may still not be perfect, but therevoid
oblique images, see Figure 17d, the result may still not be perfect, but there is much less space
is much
andvoid
less the terrain
space and modeling is more
the terrain accurate.
modeling Rossi accurate.
is more et al. [32] Rossi
confirmed
et al. that
[32] the use of aerial
confirmed that thenadiruseview
of
is not very suitable for surveying subvertical walls. On the other hand,
aerial nadir view is not very suitable for surveying subvertical walls. On the other hand, thethe contribution of oblique
images increases
contribution the consistency
of oblique of the reconstructed
images increases the consistency surfaces.
of the Tadia et al. [33]surfaces.
reconstructed improvedTadiathe vertical
et al.
accuracy of the photogrammetric model when they included oblique
[33] improved the vertical accuracy of the photogrammetric model when they included oblique images within their nadiral
dataset.
images Similarly,
within their the authors
nadiral of [34]
dataset. state that
Similarly, thethe use ofof
authors a tilted camera
[34] state thatimproves
the use ofthe robustness
a tilted cameraof
the geometrical
improves model. The
the robustness usegeometrical
of the of oblique images
model.isThe crucial
use ofto improve the density
oblique images of thetopoint
is crucial cloud,
improve
especially in connection to peculiar features of the surveyed object [35].
the density of the point cloud, especially in connection to peculiar features of the surveyed object [35].
Remote Sens. 2020, 12, 1293 16 of 20

Remote Sens. 2020, 12, x FOR PEER REVIEW 16 of 20

(a) (b)

(c) (d)

Figure 17. Dense

Figure17. Dense point
point clouds
clouds of
of aa rugged
rugged terrain
terrainand
andoverhangs,
overhangs,generated
generatedfrom:
from:(a)
(a)vertical
verticalimages
images
above
above NDTM; (b) oblique images above the fake DEM; (c) oblique images above the plane DEMand
NDTM; (b) oblique images above the fake DEM; (c) oblique images above the plane DEM and
(d)
(d)merged
mergeddense
densepoint
pointcloud
cloudfrom
fromvertical
verticaland
andoblique
obliqueimages
imagesabove
abovethe
theplane
planeDEM.
DEM.

There
Thereare
arealso
alsosome
someissues
issueswith
withthe theoblique
obliqueimages.
images. IfIf any
any objects
objects with
with significant
significantheight
heightabove
above
the
the ground
ground are present (e.g.,
are present (e.g., vegetation),
vegetation),they theycould
couldcause
causea disturbance
a disturbancein in imaging
imaging because
because theythey
are
are captured from the side, not from above as in the vertical imagery. As a result, it
captured from the side, not from above as in the vertical imagery. As a result, it is very difficult to is very difficult
to generate
generate ananorthophoto
orthophotofrom fromthetheoblique
obliqueimages
imagesifif such
such isis required.
required. Vertical
Vertical images
images areare more
more
appropriate to produce orthophotos [36]. As an example, Figure 18 shows two inserts
appropriate to produce orthophotos [36]. As an example, Figure 18 shows two inserts of orthophotos of orthophotos
of
ofaaforest
forestarea
areanext
nexttotothe
therockfall.
rockfall. ItIt is
is very
very obvious
obvious that
that the
the software
software has
has problems
problems to to create
create aa clear
clear
orthophoto
orthophoto from the oblique images. Amrullah et al. [37] suggested that only vertical imagesshould
from the oblique images. Amrullah et al. [37] suggested that only vertical images should
be
beused
usedtotoproduce
produceorthophotos.
orthophotos.
Remote Sens. 2020, 12, 1293 17 of 20
Remote Sens. 2020, 12, x FOR PEER REVIEW 17 of 20

(a) (b)
Figure 18.
Figure 18. Inserts of of
Inserts orthophoto with
orthophoto 5-cm
with resolution:
5-cm (a) from
resolution: (a) vertical imagesimages
from vertical and (b)and
from(b)
oblique
from
images. images.
oblique

The main problem with vertical images in steep terrain is the large difference in scale in a single
image, see Figures 2–4,
2-4, which probably causes longer processing times. Not only the scale is an an issue,
issue,
a big factor is the appearance
the appearance of a detail in many images.
detail in many images. For example, a detail in the lower part of
the rockfall appears in more than
than 100
100 vertical
vertical images.
images. The number of oblique images with the same
detail was about 10. In figures, a dense cloud generation from 218 vertical images took 10 hours and
54 minutes
minutesandand78.3 million
78.3 points
million were
points calculated.
were A dense
calculated. cloudcloud
A dense calculation from 203
calculation oblique
from images
203 oblique
took 1 hour
images tookand 24 minutes,
1 hour in whichin129.8
and 24 minutes, which million
129.8 points
millionwere calculated.
points were calculated.

5.3. Height Decrease

5.3. Height Decrease
An
An unexpected
unexpected issue
issue occurred
occurred onon all
all our
our flights,
flights, namely
namely thethe decrease
decrease ofof the
the actual
actual height
height of
of the
the
UAV above any DEM as the height ATO increased. At a total height of about 300 meters
UAV above any DEM as the height ATO increased. At a total height of about 300 meters ATO, the ATO, the actual
height AGL decreased
actual height by about
AGL decreased by20-25 meters
about 20-25atmeters
the top. We were
at the ablewere
top. We to compensate for this byfor
able to compensate tilting
this
the DEMs upwards, see Section 2.4.3. These values applied to the Belca site. This issue needs
by tilting the DEMs upwards, see Section 2.4.3. These values applied to the Belca site. This issue needs to be
further investigated at other sites at different elevations.
to be further investigated at other sites at different elevations.

6. Conclusions
6. Conclusions
The execution of a UAV mission in steep terrain is much more complex than in flat terrain. In the
The execution of a UAV mission in steep terrain is much more complex than in flat terrain. In
latter, the UAV flies at a constant height ATO and the images are captured vertically. In non-flat terrain,
the latter, the UAV flies at a constant height ATO and the images are captured vertically. In non-flat
the UAV must follow the terrain at a certain height AGL. Mission flying in steep terrain is much more
terrain, the UAV must follow the terrain at a certain height AGL. Mission flying in steep terrain is
comfortable than manual flying, mainly because of the changing elevation of the terrain, but still
much more comfortable than manual flying, mainly because of the changing elevation of the terrain,
special care is needed during preparation and even more during the flights. In addition to the terrain
but still special care is needed during preparation and even more during the flights. In addition to
itself, high trees or other tall objects need to be taken into account when setting the flight height.
the terrain itself, high trees or other tall objects need to be taken into account when setting the flight
To be able to follow the terrain, a DEM with a suitable resolution and topographical accuracy
height.
is required. Using low resolution global DEMs, such as SRTM, is not only inadequate, but can also
To be able to follow the terrain, a DEM with a suitable resolution and topographical accuracy is
be dangerous. Based on our research, a 5-meter NDTM contained sufficient topographic detail to
required. Using low resolution global DEMs, such as SRTM, is not only inadequate, but can also be
adequately follow the terrain from a UAV height of 80 meters. A special application is required for the
dangerous. Based on our research, a 5-meter NDTM contained sufficient topographic detail to
use of a DEM for terrain-following flying.
adequately follow the terrain from a UAV height of 80 meters. A special application is required for
A UAV survey of a steep terrain should not contain only vertical images, because there are large
the use of a DEM for terrain-following flying.
scale changes in the images and not all details are captured in the rugged terrain. The processing time
A UAV survey of a steep terrain should not contain only vertical images, because there are large
of the vertical images is much longer than the processing of the oblique images. The oblique imagery
scale changes in the images and not all details are captured in the rugged terrain. The processing time
also
of thehas an issue
vertical generating
images is much anlonger
orthophoto image.
than the processing of the oblique images. The oblique imagery
also has an issue generating an orthophoto image. the optimal solution is to use a combination of
To overcome the problems mentioned above,
vertical
To and obliquethe
overcome images. It is true
problems that the above,
mentioned combination of bothsolution
the optimal types of is
images
to userequires even more
a combination of
vertical and oblique images. It is true that the combination of both types of images requires even more
Remote Sens. 2020, 12, 1293 18 of 20

processing time, but the result is a point cloud with much more detail, and a proper orthophoto can
be produced.
Regular DTMs as a basis for the terrain following flying are only suitable for the vertical images.
Custom DEMs can be generated to create a surface for the oblique images. The plane DEM is easy
to construct, it assures the minimum distance to the surface and thus a sufficient overlap. However,
the fake DEM is preferred, as the DCGs are preserved and thus the GSD is kept constant.
If due to limited time or UAV batteries only one flight mission can be performed and the orthophoto
is not required, we recommend the use of the fake DEM and oblique images. For the optimal complete
solution, select NDTM for the vertical images and the fake DEM for the oblique images. If there is
a height issue as we encountered it, use a tilted NDTM for the vertical images and a tilted fake DEM
for the oblique images. The plane DEM is a reasonable alternative to the fake DEM.

Author Contributions: Conceptualization, K.K.T., D.G. and D.P.; Data curation, K.K.T. and D.P.; Formal
analysis, K.K.T. and D.G.; Funding acquisition, D.P.; Investigation, D.G. and D.P.; Methodology, K.K.T.;
Project administration, D.P.; Resources, D.P.; Supervision, D.P.; Validation, D.G. and D.P.; Visualization, K.K.T.;
Writing—original draft, K.K.T.; Writing—review and editing, K.K.T., D.G. and D.P. All authors have read and
agreed to the published version of the manuscript.
Funding: The authors acknowledge the financial support from the Slovenian Research Agency (research core
funding No. P2-0227 Geoinformation infrastructure and sustainable spatial development of Slovenia and No.
P2-0406 Earth observation and geoinformatics).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Mikoš, M.; Majes, B. Mitigation of large landslides and debris flows in Slovenia. In Landslides: Causes,
Types and Effects; Werner, E.D., Friedman, H.P., Eds.; Nova Science Publishers Inc.: New York, NY, USA, 2010;
pp. 105–131, ISBN 978-1-60741-258-8.
2. Skrzypczak, I.; Kogut, J.; Kokoszka, W.; Zientek, D. Monitoring of Landslide Areas with the Use of
Contemporary Methods of Measuring and Mapping. Civil. Environ. Eng. Rep. 2017, 24, 69–82. [CrossRef]
3. Kasperski, J.; Delacourt, C.; Allemand, P.; Potherat, P.; Jaud, M.; Varrel, E. Application of a Terrestrial Laser
Scanner (TLS) to the Study of the Séchilienne Landslide (Isère, France). Remote Sens. 2010, 2, 2785–2802.
[CrossRef]
4. Tofani, V.; Segoni, S.; Agostini, A.; Catani, F.; Casagli, N. Technical Note: Use of remote sensing for landslide
studies in Europe. Nat. Hazards Earth Syst. Sci. 2013, 13, 299–309. [CrossRef]
5. Tahar, K.N.; Ahmad, A.; Akib, W.A.A.W.M.; Mohd, W.M.N.W. A New Approach on Production of Slope
Map Using Autonomous Unmanned Aerial Vehicle. Int. J. Phys. Sci. 2012, 7, 5678–5686. [CrossRef]
6. Huang, H.; Long, J.; Lin, H.; Zhang, L.; Yi, W.; Lei, B. Unmanned aerial vehicle based remote sensing method
for monitoring a steep mountainous slope in the Three Gorges Reservoir, China. Earth Sci. Inform. 2017, 10,
287–301. [CrossRef]
7. Manconi, A.; Ziegler, M.; Blöchliger, T.; Wolter, A. Technical note: Optimization of unmanned aerial vehicles
flight planning in steep terrains. Int. J. Remote Sens. 2019, 40, 2483–2492. [CrossRef]
8. Westoby, M.J.; Brasington, J.; Glasser, N.F.; Hambrey, M.J.; Reynolds, J.M. ‘Structure-from-Motion’
photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology 2012, 179, 300–314.
[CrossRef]
9. Pepe, M.; Fregonese, L.; Scaioni, M. Planning airborne photogrammetry and remote-sensing missions with
modern platforms and sensors. Eur. J. Remote Sens. 2018, 51, 412–435. [CrossRef]
10. Fernández, T.; Pérez, J.; Cardenal, J.; Gómez, J.; Colomo, C.; Delgado, J. Analysis of Landslide Evolution
Affecting Olive Groves Using UAV and Photogrammetric Techniques. Remote Sens. 2016, 8, 837. [CrossRef]
11. De Beni, E.; Cantarero, M.; Messina, A. UAVs for volcano monitoring: A new approach applied on an active
lava flow on Mt. Etna (Italy), during the 27 February–02 March 2017 eruption. J. Volcanol. Geother. Res. 2019,
369, 250–262. [CrossRef]
12. Niethammer, U.; James, M.R.; Rothmund, S.; Travelletti, J.; Joswig, M. UAV-based remote sensing of the
Super-Sauze landslide: Evaluation and results. Eng. Geol. 2012, 128, 2–11. [CrossRef]
Remote Sens. 2020, 12, 1293 19 of 20

13. Möllney, M.; Kremer, J. Contour Flying for Airborne Data Acquisition. Available online: https://phowo.ifp.
uni-stuttgart.de/publications/phowo13/130Kremer.pdf (accessed on 28 March 2020).
14. Al-Rawabdeh, A.; Moussa, A.; Foroutan, M.; El-Sheimy, N.; Habib, A. Time Series UAV Image-Based Point
Clouds for Landslide Progression Evaluation Applications. Sensors 2017, 17, 2378. [CrossRef]
15. Rossini, M.; Di Mauro, B.; Garzonio, R.; Baccolo, G.; Cavallini, G.; Mattavelli, M.; De Amicis, M.; Colombo, R.
Rapid melting dynamics of an alpine glacier with repeated UAV photogrammetry. Geomorphology 2018, 304,
159–172. [CrossRef]
16. Cook, K.L. An evaluation of the effectiveness of low-cost UAVs and structure from motion for geomorphic
change detection. Geomorphology 2017, 278, 195–208. [CrossRef]
17. Valkaniotis, S.; Papathanassiou, G.; Ganas, A. Mapping an earthquake-induced landslide based on UAV
imagery; case study of the 2015 Okeanos landslide, Lefkada, Greece. Eng. Geol. 2018, 245, 141–152. [CrossRef]
18. Agüera-Vega, F.; Carvajal-Ramírez, F.; Martínez-Carricondo, P.; Sánchez-Hermosilla López, J.;
Mesas-Carrascosa, F.J.; García-Ferrer, A.; Pérez-Porras, F.J. Reconstruction of extreme topography from UAV
structure from motion photogrammetry. Measurement 2018, 121, 127–138. [CrossRef]
19. Pellicani, R.; Argentiero, I.; Manzari, P.; Spilotro, G.; Marzo, C.; Ermini, R.; Apollonio, C. UAV and
Airborne LiDAR Data for Interpreting Kinematic Evolution of Landslide Movements: The Case Study of the
Montescaglioso Landslide (Southern Italy). Geosciences 2019, 9, 248. [CrossRef]
20. Sarro, R.; Riquelme, A.; García-Davalillo, J.; Mateos, R.; Tomás, R.; Pastor, J.; Cano, M.; Herrera, G. Rockfall
Simulation Based on UAV Photogrammetry Data Obtained during an Emergency Declaration: Application
at a Cultural Heritage Site. Remote Sens. 2018, 10, 1923. [CrossRef]
21. Lucieer, A.; Jong, S.M.; de Turner, D. Mapping landslide displacements using Structure from Motion (SfM)
and image correlation of multi-temporal UAV photography. Prog. Phys. Geogr. 2014, 38, 97–116. [CrossRef]
22. Yang, C.-H.; Tsai, M.-H.; Kang, S.-C.; Hung, C.-Y. UAV path planning method for digital terrain model
reconstruction—A debris fan example. Autom. Constr. 2018, 93, 214–230. [CrossRef]
23. Mikhail, E.M.; Bethel, J.S.; McGlone, J.C. Introduction to Modern Photogrammetry; John Wiley & Sons: New York,
NY, USA, 2001; pp. 30–31, ISBN 978-0-471-30924-6.
24. Shuttle Radar Topography Mission. Available online: https://www2.jpl.nasa.gov/srtm/ (accessed on 11
October 2019).
25. Terrain Following. Available online: https://ardupilot.org/plane/docs/common-terrain-following.html
(accessed on 27 February 2020).
26. Kolecka, N.; Kozak, J. Assessment of the Accuracy of SRTM C- and X-Band High Mountain Elevation Data:
A Case Study of the Polish Tatra Mountains. Pure Appl. Geophys. 2014, 171, 897–912. [CrossRef]
27. Tahar, K.N. An evaluation on different number of ground control points in unmanned aerial vehicle
photogrammetric block. ISPRS—Int. Arch. Photogram. Remote Sens. Spat. Inform. Sci. 2013, XL-2/W2, 93–98.
[CrossRef]
28. Nasrullah, A.R. Systematic Analysis of Unmanned Aerial Vehicle (UAV) Derived Product Quality. Master’s
Thesis, Faculty of Geo-Information Science and Earth Observation of the University of Twente, Enschede,
The Netherlands, 2016.
29. Agüera-Vega, F.; Carvajal-Ramírez, F.; Martínez-Carricondo, P. Accuracy of Digital Surface Models and
Orthophotos Derived from Unmanned Aerial Vehicle Photogrammetry. J. Surv. Eng. 2017, 143, 04016025.
[CrossRef]
30. Agisoft Metashape. Available online: https://www.agisoft.com (accessed on 21 September 2019).
31. Volinsky, H. Personal e-mail Communication; Drone Harmony: Lucerne, Switzerland, 2018.
32. Rossi, P.; Mancini, F.; Dubbini, M.; Mazzone, F.; Capra, A. Combining nadir and oblique UAV imagery to
reconstruct quarry topography: Methodology and feasibility analysis. Eur. J. Remote Sens. 2017, 50, 211–221.
[CrossRef]
33. Taddia, Y.; Stecchi, F.; Pellegrinelli, A. Coastal Mapping using DJI Phantom 4 RTK in Post-Processing
Kinematic Mode. Drones 2020, 4, 9. [CrossRef]
34. Manfreda, S.; Dvorak, P.; Mullerova, J.; Herban, S.; Vuono, P.; Arranz Justel, J.; Perks, M. Assessing the
Accuracy of Digital Surface Models Derived from Optical Imagery Acquired with Unmanned Aerial Systems.
Drones 2019, 3, 15. [CrossRef]
Remote Sens. 2020, 12, 1293 20 of 20

35. Chiabrando, F.; Lingua, A.; Maschio, P.; Teppati Losè, L. The Influence Of Flight Planning And Camera
Orientation In Uavs Photogrammetry. A Test In The Area Of Rocca San Silvestro (Li), Tuscany. Int. Arch.
Photogramm. Remote Sens. Spat. Inf. Sci. 2017, XLII-2/W3, 163–170. [CrossRef]
36. Seier, G.; Sulzer, W.; Lindbichler, P.; Gspurning, J.; Hermann, S.; Konrad, H.M.; Irlinger, G.; Adelwöhrer, R.
Contribution of UAS to the monitoring at the Lärchberg-Galgenwald landslide (Austria). Int. J. Remote Sens.
2018, 39, 5522–5549. [CrossRef]
37. Amrullah, C.; Suwardhi, D.; Meilano, I. Product Accuracy Effect Of Oblique And Vertical Non-Metric Digital
Camera Utilization In Uav-Photogrammetry To Determine Fault Plan E. ISPRS Ann. Photogramm. Remote
Sens. Spat. Inf. Sci. 2016, III–6, 41–48. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).