You are on page 1of 81

University of Miami

Scholarly Repository
Open Access Theses

Electronic Theses and Dissertations

2014-11-29

Using Unmanned Aerial Vehicles and "Structure


from Motion" Software to Monitor Coastal Erosion
in Southeast Florida
Conor Maguire
University of Miami, cmaguire@rsmas.miami.edu

Follow this and additional works at: http://scholarlyrepository.miami.edu/oa_theses


Recommended Citation
Maguire, Conor, "Using Unmanned Aerial Vehicles and "Structure from Motion" Software to Monitor Coastal Erosion in Southeast
Florida" (2014). Open Access Theses. Paper 525.

This Open access is brought to you for free and open access by the Electronic Theses and Dissertations at Scholarly Repository. It has been accepted for
inclusion in Open Access Theses by an authorized administrator of Scholarly Repository. For more information, please contact
repository.library@miami.edu.

UNIVERSITY OF MIAMI

USING UNMANNED AERIAL VEHICLES AND STRUCTURE FROM MOTION


SOFTWARE TO MONITOR COASTAL EROSION IN SOUTHEAST FLORIDA

By
Conor Maguire
A THESIS

Submitted to the Faculty


of the University of Miami
in partial fulfillment of the requirements for
the degree of Master of Science

Coral Gables, Florida


December 2014

2014
Conor Maguire
All Rights Reserved

UNIVERSITY OF MIAMI

A thesis submitted in partial fulfillment of


the requirements for the degree of
Master of Science

USING UNMANNED AERIAL VEHICLES AND STRUCTURE FROM MOTION


SOFTWARE TO MONITOR COASTAL EROSION IN SOUTHEAST FLORIDA
Conor Maguire
Approved:
________________
John McManus, Ph.D.
Professor of Marine Biology
And Fisheries

_________________
Maria Estevanez, Ph.D.
Sr. Lecturer, Marine Affairs
and Policy

________________
Manoj Shivlani, Ph.D.
Adjunct Professor, Marine Affairs
and Policy

_________________
M. Brian Blake, Ph.D.
Dean of the Graduate School

MAGUIRE, CONOR
Using Unmanned Aerial Vehicles and
Structure from Motion Software to
Monitor Coastal Erosion in Southeast Florida

(M.S., Marine Affairs and Policy)


(December 2014)

Abstract of a thesis at the University of Miami.


Thesis supervised by Professors John McManus and Maria Estevanez.
No. of pages in text. (71)
Coastal Erosion poses a significant threat to several coastal communities in
Florida. Ecosystem managers are tasked with both protecting the community from storm
events and preserving the natural ecosystems found in the coastal zone. The most widely
utilized method for shoreline protection in Florida is beach renourishment. Beach
renourishment projects are difficult to track with regularity because the methods used to
monitor and measure coastal erosion are cost-prohibitive. Current methods of coastal
erosion tracking include Terrestrial Laser Scanning, airborne- Light Detection and
Ranging (LiDAR) scans, and Real Time Kinematic Global Positioning Surveys. These
methods are time consuming and resource intensive. Unmanned Aerial Vehicles are
gaining popularity amongst both the public and researchers, and could offer a cheaper
alternative to airborne- LiDAR. Coupled with Structure from Motion software, UAVs
can track coastal erosion along stretches of beach on a more frequent basis. UAVs are
also better suited to react to storm events to investigate how the coast was impacted in the
immediate aftermath. This thesis found that UAVs are both feasible and cost-effective
when used in this context. The major limiting factor involved was the GPS device used.
Structure from Motion technology proved to be accurate within this context. Agisoft
PhotoScan Pro was used to develop 3 dimensional models of the beach ecosystem

scene. Digital Elevation Models (DEM) were then generated using z-values and known
positions of Ground Control Points (GCP). The geo-referenced DEM was then imported
into ESRIs AcrGIS, where it was compared to previously-generated DEMs using the
Raster Calculator tool. This method allowed for easy visualization of erosion along the
coastline. While the data proved to be useful in qualitative analysis, the accuracy of the
GPS used in this study prevents the results from being used in a quantitative function.
With accurate GPS, such as Real Time Kinematic GPS which has sub-centimeter level
accuracy, this method could produce survey-grade DEMs and be a feasible alternative to
LiDAR.

TABLE OF CONTENTS
Page
LIST OF FIGURES .....................................................................................................

iv

LIST OF TABLES .......................................................................................................

vi

Chapter
1

INTRODUCTION .........................................................................................
1.1 Background
1.2 Previous Research...
1.3 Study Area..

1
2
4
6

EQUIPMENT ................................................................................................
2.1 3D Robotics Y6 Hexacopter ......................................................................
2.2 Sony NEX-5R ............................................................................................
2.3 Camera Mounting System..........................................................................
2.4 Computer Processing .................................................................................

10
10
13
14
16

METHODOLOGY ..........................................................................................
3.1 Mission Planning .......................................................................................
3.2 Site Preparation ..........................................................................................
3.3 Flight Preparations .....................................................................................
3.4 Data Processing..........................................................................................

18
20
29
34
41

RESULTS ........................................................................................................

47

DISCUSSION ..................................................................................................
5.1 Cost ............................................................................................................
5.2 Future Research .........................................................................................
5.3 Applications ...............................................................................................

61
65
66
67

CONCLUSIONS..............................................................................................

68

WORKS CITED ......................................................................................

70

iii

LIST OF FIGURES
Figure 1.0 Jim Houstons Beach Width versus Damage to Structures
Findings..2
Figure 1.1 Project Location.....7
Figure 1.2 Atlantic Dunes Google Earth Orthophoto....................................................7
Figure 1.3 Florida Geological Survey map of Palm Beach County and Atlantic Dunes
Beach...8
Figure 2.1 3D Robotics Y6 Airframe....11
Figure 2.2 Camera Mounting System....15
Figure 3.0 Overview of Methodology...18
Figure 3.1 Flow Chart depicting Mission Planning Phase Workflow...20
Figure 3.2 Mission Planners Flight Plan Window with Google Earth Interface......23
Figure 3.3 Bounding Box around the Area to be Photographed24
Figure 3.4 The Auto-Waypoint Interface...24
Figure 3.5 Mission Planner Footprints....25
Figure 3.6 Diagram of Nadir Point27
Figure 3.7 Extrapolated Digital Elevation Model Examples.28
Figure 3.8 Ground Control Point Design...31
Figure 3.9 Ground Control Points displayed in Agisoft PhotoScan..33
Figure 3.10 June 25th, 2014 Ground Control Points in Aerial Photograph..33
Figure 3.11 Mission Planners Flight Screen.40
Figure 3.12 Conceptual Map of Data Processing Phase41
Figure 4.0 Profile Line Graphs for May 22, 2014.49
Figure 4.1 Profile Line Graphs for June 25, 2014.49
Figure 4.2 Profile Line Graphs for September 19, 201450
Figure 4.3 Elevation Changes between September and May51
Figure 4.4 May 22, 2014 Digital Elevation Model52
Figure 4.5 June 25, 2014 Digital Elevation Model53
Figure 4.6 September 19, 2014 Digital Elevation Model..54
Figure 4.7 May 22, 2014 Ground Control Points..55
iv

Figure 4.8 June 25, 2014 Ground Control Points..56


Figure 4.9 September 19, 2014 Ground Control Points.57
Figure 4.10 May Survey Data58
Figure 4.11 June Survey Data59
Figure 4.12 September Survey Data..60
Figure 5.0 PhotoScan Project depicting Camera Locations and Angles...63

LIST OF TABLES
Table 2.0 Weight Values for Independent Components.13
Table 2.1 Minimum and Recommended System Requirements.....16
Table 2.2 NovaBench Score for Alienware 14...........................................................17
Table 3.0 June 25, 2014 Ground Control Point Table with X, Y, and Z Values34
Table 5.0 Total Investment in System.....66

vi

CHAPTER 1: INTRODUCTION
Several methods of erosion monitoring currently exist and are being utilized by
municipalities and organizations. These include airborne-Light Detection And Ranging
(LiDAR) surveys, ground-based LiDAR surveys, manual beach profiling surveys, Global
Navigation Satellite System (GNSS) surveys, and photogrammetric surveys. All of these
survey methods involve highly-skilled technicians, expensive equipment, and extensive
planning. With the advent of open-source software and cheap computer processers such
as Arduino and Beagle Board, hobbyists and select companies have developed easy-to
use autopilot systems designed for traditional radio-controlled airplanes, helicopters, and
multi-rotor vehicles. This has led to exponential growth in the industry and has ushered in
new products that are ideally suited to take on the task of coastal erosion monitoring.
Now equipped with fully-functional autopilots, these radio-controlled vehicles can be
tasked with surveying large areas and hard to reach areas, all without minimal user input.
Airframes such as 3DRobotics Y-6 hexacopter are capable of lifting DSLR cameras into
positions that mimic the use of airplanes and helicopters for surveying. Using computervision and Structure from Motion software, a user can simply photograph an area with
a relatively cheap Unmanned Aerial Vehicles (UAV), load the photos into a software
program, and produce highly detailed Orthophotos, Digital Elevation Models, and 3D
Models. The current leader in Structure from Motion software is Agisofts PhotoScan
Pro. This program utilizes the programming from Autodesks AutoCAD and is easily
integrated into workflows with both AutoCAD/Map3D and ESRIs ArcGIS, the
standard in environmental GIS software.

2
1.1 BACKGROUND
Coastal Erosion is a significant threat to major cities, tourism areas, and
residences; particularly in Florida. The beach ecosystem relies upon the transfer and
deposition of sand along the coast to naturally renourish areas while other areas
experience erosion. There are several causes of erosion of a beach ecosystem, both
natural and manmade (Ruppert, 2008). Natural forces of erosion include storms, waves,
and strong currents. Erosion can also be influenced by people. Examples of humaninfluenced erosion include blocking sediment transfer due to inlets and jetties,
significantly changing the coastline through fills, and armoring coastal areas (Ruppert,
2008). These forces are all at play on the East Coast of Florida.
The impacts of coastal erosion in Florida are two-fold. First, the beach ecosystem
acts as a strong natural barrier to storm surge. When the beach is eroded critically, it no
longer offers sufficient protection from the storms impact (Ruppert, 2008). Jim Houston,
of the Army Corps of Engineers found that structural damage exponentially increased as
beach width decreased (Houston, 2013).

Figure 1.0 Jim Houstons beach width versus damage to structures findings.
The second aspect of coastal erosion in Florida regards tourism. With a small
beach, fewer beach-goers will visit. Studies have shown an increase in beach attendance

3
is proportional to beach width (Houston, 2013). It could be argued that severe erosion
would not only hurt the ability of the coast to mitigate storm damage, but also damage the
economy of the coastal region through decreased beach attendance revenue.
The Florida Department of Environmental Protection (FLDEP) tracks coastal
erosion in Florida and has built up an extensive database to record severe erosion and
erosion events throughout the state. According to the FLDEP, approximately 5% of
Floridas shoreline is critically eroding. Of this area, 398.6 miles of beach was
experiencing critical erosion in 2010. This figure was an increase over the previous figure
of 327.9 miles of critically eroded beach in 2000. As FLDEP states, of the 825 miles of
sandy beaches in Florida, 494.5 miles are considered either critically or non-critically
eroded (FLDEP, 2013). In Florida, the definition of critically eroded is a segment of
shoreline where natural processes of human activities have caused or contributed to
erosion and recession of the coastal system to such a degree that upland development,
recreation, wildlife habitat, or important cultural resources are threatened or lost
(FLDEP, 2013).
Monitoring coastal erosion is both difficult and time consuming. Traditional
methods require skilled technicians, expensive equipment, and a large investment. For
these reasons, monitoring projects are usually carried out only by governments, or
contractors working under a government grant. Local municipalities often do not have the
time, resources, or money to monitor all of the coastal areas under their control.
Additionally, coastal monitoring is typically only cost effective when done on a large
scale. Smaller projects simply do not match the economies of scale that a large regional
project would provide for technologies such as standard LiDAR. For this reason, small

4
sections of coastline can often be ignored by a coastal monitoring project. To rectify the
situation, coastal monitoring tools need to be cheap, easy to use, and able to accurately
map small areas of coastline.
In recent years, Unmanned Aerial Vehicles have experienced an exponential
growth in development. Originally used in military applications, UAV are now being
used in very diverse roles. Conservation Drones uses drones to map wildlife, detect
poachers, and increase the general knowledge of endangered species through the use of
UAVs (Drones, 2014). 3D Robotics uses UAV to assist in vineyard management in
California, and some UAV are now being used for precision agriculture (Paskulin, 2013).
The many uses of UAV and their popularity amongst the public have combined to
generate a market that now supplies countless varieties of UAV to civilian users. In
addition, with the support of online communities, developers have created free software
for flight control, as well as hardware designed to fully automate the functions of a UAV.
The increased use has also significantly driven the price of UAV down. For these
reasons, UAV are an ideal platform to monitor coastal erosion.
1.2 PREVIOUS RESEARCH
The use of UAV for coastal research is a new phenomenon. Previously, as
mentioned earlier, erosion studies depended on airborne- LiDAR or Terrestrial Laser
Scanning to gather enough information to generate a digital elevation model. These
methods were both time-consuming and cost-prohibitive for many municipalities and
researchers. With the advent of open-source UAV systems and the development of
stronger computer vision models, such as Structure from Motion software, the ability to
use small UAV has significantly increased. Several research projects have investigated

5
the use of Structure from Motion technology as a means to generate digital elevation
models. In addition, several research projects have also investigated the use of UAV for
remote sensing in lieu of satellites or manned aircraft. Mancini et. Al (2013) utilized a
multi-rotor airframe, coupled with Agisofts PhotoScan Pro software to generate digital
elevation models of a beach ecosystem in Italy. One study compared the digital elevation
model to one generated by a Terrestrial Laser Scanner (TLS) survey (Mancini, et al.,
2013). To do this, a Networked Real Time Kinematic Global Positioning System (NRTK
GPS) was used to locate a set of Ground Control Points for both methods. The NRTK
GPS system allowed for centimeter-level precision in order to accurately compare each
method. The study found that the Structure from Motion technique more closely
represented the NRTK survey values than the TLS survey (Mancini, et al., 2013). This
study validated the use of PhotoScan Pros Structure from Motion software as a costeffective alternative to TLS surveys, assuming certain conditions were met.
In addition to the Mancini Study results, another study investigated the use of
UAV for digital elevation modeling using a separate Structure from Motion algorithm
(Eisenbeiss & Zhang, 2006). Eisenbess and Zhang found that the Structure from
Motion software and the Terrestrial Laser Scanner model differed by only 6 cm, which
could be explained occlusion, a phenomenon in which surfaces or points are not visible in
a particular viewpoint (Eisenbeiss & Zhang, 2006). Both papers establish Structure from
Motion as a usable alternative method to traditional digital elevation model generation
techniques. A separate study used a UAV and Structure from Motion software utilizing
multi-view stereo image algorithms to map Antarctic moss beds in high definition
(Lucieer, Robinson, Turner, Harwin, & Kelcey, 2012). The study reported a resolution of

6
2 cm using the Structure from Motion software (Lucieer, Robinson, Turner, Harwin, &
Kelcey, 2012). Finally, another study used a UAV and Structure from Motion software
to develop digital elevation models using both Ground Control Points and no Ground
Control Points (Turner, Lucieer, & Watson, 2012). The study found that the digital
elevation models using only camera coordinates generated an accuracy of 65-120 cm,
while the digital elevation models utilizing Ground Control Points generated an accuracy
of 10-15 cm (Turner, Lucieer, & Watson, 2012).
As shown, previous research has established the accuracy of Structure from
Motion software and UAV based digital elevation models. However, it is important to
note the common factors between each study necessary for proper digital elevation
modelling. First, Ground Control Points were used in every study. As the Turner et al
(2012) paper demonstrated, failure to use Ground Control Points resulted in accuracies 6
to 10 times less than projects that do utilize Ground Control Points. Second, the use of an
accurate GPS device, namely a Network Real Time Kinematic GPS device or similar
GNSS device capable of receiving both L1 and L2 signals from GPS satellites, is
required if the resulting digital elevation model is to be used in a survey-grade function.
1.3: STUDY AREA
Delray Beach is a coastal community located on the Eastern coast of Florida.
Geographically, Delray Beach is located at {26.459167, -80.083056}. The elevation of
Delray Beach averages five meters, making it extremely low-lying. Much of the coast is
at or slightly above sea level. Coastal flooding is common across the immediate coastal
zone during high tides, storms, and hurricanes. Delray Beach is in a location frequented
by hurricanes and powerful winter storms and swells. Delray Beachs coastline is roughly

7
4.28 kilometers long and runs in a North-South orientation. There is very little deviation
from this orientation and no bays, harbors, or man-made inlets exist within the city limits
of Delray Beach.

Figure 1.1 Project Location. Source: http://pics2.citydata.com/city/maps/fr506.png

Figure 1.2 Atlantic Dunes Beach, Delray Beach, Fl. Source: Google Earth

8
The coastal ecosystem of Delray Beach is that of sandy beach transitioning into
dunes. The majority of the beach is public property, and the Coastal Construction Control
Line has been established west of the State Road A1A, a rare occurrence on the densely
populated Southeast Coast of Florida. For this reason, the dune ecosystem has been given
a chance to move, replenish, and recover from earlier development east of A1A. The
dunes average 50 meters wide and stretch from the north end of the Delray Municipal
Beach to the south end of the beach. While a small seawall exists west of the dune
systems, the principal erosion control method utilized by Delray Beach is beach
renourishment. The primary geological features of Atlantic Dunes Beach Park are
Anastasia formations with lithified coquina of shells and sand and unlithified
fossiliferous sand, according to the Florida Geological Survey referenced

below.
Figure 1.3 Florida Geological Survey map for Palm Beach County and Atlantic
Dunes Beach (located north of Highland Beach and south of Ocean Ridge (Survey,
2014))
Delray Beach has a long history of beach renourishment. The roughly 4
kilometers of coastline have been renourished a total of seven times, including most

9
recently in 2013 (Little, 2014). To date, the renourishment has been considered
successful. Prior to renourishment, almost no beach remained, due to the combination of
excessive erosion and a seawall. In 1974, the City began renourishing the beach, as well
as replanting and restoring the dune ecosystem (Little, 2014). The first successful
renourishment was followed by subsequent renourishments conducted in 1978, 1984,
1992, 2002, 2005, and 2013. The 10 year cycle of renourishment has been disrupted
twice by hurricanes, first in 2005 to replenish the beach after a series of hurricanes passed
through the area, and again in 2013 after time-related erosion and Hurricane Sandy
impacted the coast. This roughly 10 year cycle was determined to be the appropriate time
frame for a beach renourishment project due to the large cost of such a project and the
estimated time frame for which it would take the beach to erode to a certain point. This
cycle is very important to the City, as 51% of beach front is considered public. In
addition, the economic benefits of the public beach are numerous. A study conducted in
1995 by the City found that the beach renourishment project increased property values by
roughly $228.8 million in the surrounding area (Little, 2014).

CHAPTER 2: EQUIPMENT
2.1 3DRobotics Y6 Hexacopter
For this study, the 3DRobotics Y6 hexacopter was utilized as the UAV. This
UAV is a capable, customizable, fully autonomous airframe that is both versatile and
durable. The hexacopter utilizes a Y shaped airframe. This locates 2 motors on each
arm, one acting as a pusher motor and the other as a puller motor. This setup was chosen
for multiple reasons. First, if a motor fails, there is a higher probability of being able to
recover the UAV and perform a controlled landing. In a quadcopter design, the loss of a
motor is usually too catastrophic for a human pilot to compensate for quickly enough.
The over-under orientation of the motors provide a degree of stability (Robotics, 3DR
RTF Y6, 2014), and the six motors increase lift capacity substantially. In addition, the
Y formation is considerably smaller in footprint than a traditional hexacopter. The
Y6 also has the ability to be folded for ease of transport. This degree of protection
during transportation is a vital part of the utility of the Y6. The drawbacks of the Y
airframe are also worth considering. The most significant drawback of the Y airframe
in a hexacopter is the inefficiency of the motor configuration. While efforts are taken to
minimize the loss of efficiency through sizing the top and bottom propellers differently
and utilizing counter-rotating motors, the loss of efficiency on the setup has been
calculated at 25% (Coleman, 1997).
The 3DRobotics Y6 UAV comes equipped with an APM 2.6 autopilot. The
APM 2.6 is a fully capable autopilot that allows the user to pre-program mission profiles
from takeoff to landing. This key feature saves time when in the field and allows for
completely repeatable missions, ensuring the same area is covered on multiple occasions.
The APM 2.6 autopilot uses Linux coding and is built upon the Arduino board design.
10

11
This enables easy editing and customization of the module should a part fail or the
operator feels a need to upgrade. The ability to upgrade the APM, as well as the Y6
frame, allows the user to improve the performance significantly, without the need to
purchase a brand new setup.

Figure 2.1 3D Robotics Y6 airframe. Source:


https://store.3drobotics.com/products/apm-3dr-Y6-rtf

12
Modifications to the Y6 airframe included replacement of the arm mounting
bolts with longer bolts, upgrading the battery, and modifying the front deck of the
airframe to accept the bracket to hold the camera. The longer bolts were utilized in order
to provide more options for the camera suspension system. This system was comprised of
bungee cords that supported the camera, yet allowed the camera to move separately from
the airframe, reducing vibrations. The bungee cords were attached on one side to existing
holes in the airframe, and on the other side the longer bolts provided a solid and easy to
attach mounting point for the bungee cords hooks. This system worked well and took
very little time to adjust, remove, and apply. The bungee cords supported the camera
almost entirely, without transferring the vibrations to the lens. Upon fitting the camera to
the airframe for the first flight, it became clear that a 3S battery did not have the power to
support such a heavy system. The battery was upgraded from the 3S 3500 mAh battery
provided with the airframe to the Turnigy Li-Po battery with a 4S 5000 mAh capacity
and a discharge rate of 25-35C. The discharge rate was paired to the original battery in
order to prevent short-circuiting the wiring systems.
The weight of the system pushed the limits of the airframe, and more power was
needed to ensure the UAV could maintain an acceptable altitude with the camera
attached. The 4S battery accomplished this goal and allowed for longer flight times and
higher performance flying. The drawback of the 5000 mAh setup is the weight; the
battery weighed 544 grams, or approximately double the weight of the included 3S 3500
mAh battery at 265 grams. The Y6 has a payload capacity of 600 grams, enough to
carry the mount, camera, lens, and hood. The combined setup of the UAV weighed 2,140

13
grams, or 2.14 kilograms. The airframe weighed 1,262 grams, while the camera weighed
354 grams.
Table 2.0 Weight values for individual components

Component
Airframe
w/ mount
Battery
Sony NEX-5R
Camera Body
16 mm Pancake
Lens
Hood
Total

Weight
(grams)
1262
524
278
65
11
2140 grams

2.2 SONY NEX5R


The Sony Nex-5R DSLR camera is a 16 megapixel interchangeable lens camera
system that is both compact and extremely powerful. The NEX-5R combines a very large
image sensor, at 16 megapixels, with a relatively compact body. The camera also features
an interchangeable lens. These characteristics make the NEX-5R a desirable budget
camera for aerial photogrammetry. The lens used for this study was a 16 mm pancake
lens. The fixed lens prevents the camera from autofocusing, which makes point
calculations between images much simpler. Without a fixed lens, the Digital Elevation
Model would suffer in either production time or accuracy. The interchangeable lens also
provides the user with many options and the ability to carry multiple lenses. Sand in the
lens during a flight may render the lens useless, but the camera body would still be in
working condition. The interchangeable lens allows the user to quickly change out the
damaged lens for a new one, reducing the chance for a failed project due to faulty

14
equipment. The lens used in this study was also fitted with a lens hood. Due to the
downward orientation of the camera, it was necessary to block light from impacting the
image sensor directly. The lens hood achieved this objective while also providing
secondary lens protection during hard landings and crashes. The lens hood added a
negligible amount of weight and flight performance was unchanged whether the hood
was attached or off the lens. The combined weight of the camera body, lens, and lens
hood totaled 354 grams.
In addition to the benefits of the small size and high power of the camera, the
software powering the camera allowed for additional scripting. By allowing scripts to be
uploaded to the camera, the camera negated the need for an infrared or physical
intervalometer that would have added a considerable amount of weight. The camera,
through software, achieved the same functionality as the infrared trigger without the extra
components and complexity. An intervalometer script is already available for the NEX5R (Sony, 2014). It is important to note that this script is made available by Sony, and is
only compatible with the NEX-5R and NEX-5T.
2.3 CAMERA MOUNTING SYSTEM

The camera hard-mount was constructed from angled steel. The angled steel
was shaped like an L and allowed for the utilization of the cameras existing mounting
bolt for tripods, while being oriented in the correct position, as depicted in Figure 2.2. An
additional modification to the camera mount was the addition of rubber washers that
significantly reduced shocks being transmitted from the UAV to the camera. This simple
and effective modification allowed the camera to capture crisp, high resolution images
without the typical interference from the small, constant vibrations associated with small

15
UAV motors. The camera mount was fixed to the airframe using a 4mm bolt with both
steel washers and rubber washers for support and vibration isolation. The camera was
bolted to the mount through the standard DSLR tripod bolt, a size 20 bolt. The single
hard-point connection ensured the camera would not fall, should a bungee cord break.
The camera mounting system was complemented by two bungee cords cross-bracing the
camera and utilizing existing holes in the airframe as well as a modified arm bolt.

Figure 2.2 Camera mounting system

16
2.4 COMPUTER PROCESSING
In order to process the large file sizes and complex operations required for this
study, a reasonably high-performance computer was necessary. As all programs run
natively in a Windows environment, a Windows-based laptop was chosen. The
requirements for the Agisoft PhotoScan Pro and ArcGIS programs are detailed
below:
Table 2.1 Minimum and Recommended System Requirements
Minimum System Requirements
OS
CPU
Speed
XP or
PhotoScan Pro Later
Intel Core 2 Duo
2.2 GHz
ArcGIS
XP or
Intel Core Duo,
10.2
Later
SSE2
2.2 GHz
Recommended System Configuration
OS
CPU
Speed
Windows
Open CL
PhotoScan Pro 7
Intel Core i7
supported
HyperThreading
ArcGIS
Windows Intel Pentium 4,
or Multi10.2
7
Xeon
core

RAM
2 GB
2 GB
RAM
12
GB

2 GB

The computer chosen for this study was an Alienware 14 laptop. This laptop ran
Windows 8, with downgrade rights to Windows 7 Pro, allowing for all programs to be
run on the laptop without issue. The computer was built with an Intel Core i74710MQ
processor running at 2.50GHz @ 2501 MHz and up to 3.5GHz while using TurboBoost.
The computer utilized 16 GB Dual Channel DDR3L Ram at 1600MHz on two 8GB
sticks. The GPU was a NVIDIA GeForce GT 750M with 2GB of GDDR5. This setup
allowed for GPU acceleration and OpenGL processing, minimizing processing time. In

17
addition, the Intel Core i7 utilized 4 real cores, with 8 threads, which allowed for multitasking and easy switching between resource intensive programs such as ArcGIS and
Photoscan Pro. The Novabench score the configured laptop was 1147:
Table 2.2 Novabench score for Alienware 14
NovaBench Score: 1147
8/31/2014 9:10:27 PM
Microsoft Windows 8.1
Intel Core i74710MQ 2.50GHz @ 2501 MHz
Graphics Card: Intel(R) HD Graphics 4600
16266 MB System RAM (Score: 244)
- RAM Speed: 13347 MB/s
CPU Tests (Score: 751)
- Floating Point Operations/Second: 205132560
- Integer Operations/Second: 847430808
- MD5 Hashes Generated/Second: 1174143
Graphics Tests (Score: 101)
- 3D Frames Per Second: 318
Hardware Tests (Score: 51)
- Primary Partition Capacity: 455 GB
- Drive Write Speed: 188 MB/s

CHAPTER 3: METHODOLGY
In order to more efficiently use both time and materials while in the field, it was
necessary to utilize a structured approach. By developing and using a structure or
method, uniformity could be achieved, and operations spanning multiple site visits would
be more accurate when all steps in the method have been followed each time. In order to
develop the most efficient method for UAV-based digital elevation modelling, I divided
the overall method into four distinct categories: Mission Planning, Site Preparation,
Flight Operations, and Data Processing. All of these categories are components of a
larger method and cannot be singled out if accurate results are desired. Each category
engages a particular aspect of Photogrammatry, Unmanned Aerial Vehicle Operations,
and Erosion Monitoring.

Figure 3.0 Overview of Methodology


Mission Planning utilizes principles of Photogrammetry such as Nadir, Angle of
Incidence, Ground Resolution, and more. Site Preparation is based upon proven and
effective methods in both Remote Sensing and Digital Elevation Modelling. These
methods include the use of Ground Control Points, preliminary surveys, and ground-

18

19
truthing. It also entails planning flight paths and flight characteristics that are appropriate
to the site and conditions.
Site Preparation involves developing a layout for ground control points, checking the
accuracy of the satellite image used for mission planning, ensuring no obstacles are in the
path of the UAV, and evaluating environmental conditions to determine if a flight can be
safely undertaken. Site preparation is perhaps the most important step because the Digital
Elevation Model and Orthophotos depend on the proper placement of the ground control
points. It is imperative to achieve the minimum requirements for ground control point
placement. If a site is not fully encompassed by GPS coordinates obtained through a true
GPS unit on the ground, significant warping of the model is possible during data
processing.
Flight Operations relies on methods developed and practiced for decades by the
hobbyist community. It is important to fully comprehend the scope and value of the
contributions made by recreational RC communities and developers. Without them, and
open-source software, UAV-based Digital Elevation Modelling would only be possible
for large-scale corporations and large governments. Open-source software drives the
price of the entire operation down into the budgets suitable for small companies, town
governments, and researchers both working individually and as part of small institutions.
Data Processing relies upon resource-intensive computing and highly automated
scripts in order to take the images obtained during flight operations and translate the data
into a 3 dimensional digital elevation model.

20
3.1 MISSION PLANNING

Figure 3.1 Flow-Chart depicting Mission Planning phase workflow


Mission planning is perhaps the most crucial aspect to a successful operation. Mission
planning involves combining all aspects of the project and ensuring that every goal will
be met and planned for. These aspects include obtaining usable resolution images,
adequate image coverage, sufficient images for data processing, and precise enough GPS
coordinates for the intended use. It is important to note that, in DEM generation, Z-value
accuracy can be the most important factor and is generally two to three times less
accurate than the X- and Y- values obtained by the same global positioning system
(Garmin, 2005). This margin of error must be accounted for when designing a mission.
For example, if the end product of the project is a usable digital elevation model for
construction purposes, survey-grade GPS equipment such as GNSS and RTK GPS
systems would be required, as the typical Z-Values for consumer-level GPS units have a

21
margin of error of 2-3 meters (Garmin, 2005). However, if the DEM is only to be used
as a visual reference or a purpose in which trends are being monitored and not
centimeter-level changes, consumer-level GPS units should suffice, as they did in this
study. The quality and accuracy of the GPS equipment is a priority because, while the
Structure from Motion software is extremely accurate and will correctly place points
according the scale assigned, the model still relies upon ground control points generated
by the user and the GPS equipment.
Atlantic Dunes Park was chosen because of several beneficial factors. First, the park
was recently re-nourished. A recently re-nourished beach will demonstrate greater
erosion/accretion changes due to the excess amount of sand available. The looselypacked sand on the seaward side of a renourished beach is more likely to erode at a faster
rate than a beach with hard-packed sand. In addition, Atlantic Dunes is a public park,
with no privacy concerns that would be prevalent on a beach flanked by private homes.
Atlantic Dunes has a tall stand of Australian Pine trees between the beach and the main
road, A1A. These trees offer visual protection and a physical barrier that would prevent
the aerial vehicle from flying over a well-traveled highway if control of the vehicle is
lost. The trees also blocked wind flow out of the west, allowing for more suitable
conditions than an area with small or no vegetation. These factors were all taken into
account when the missions were planned.
After selecting a proper site, the Mission Planning phase of the methodology requires
careful observation of the site in situ. In order to properly plan a mission, all obstacles,
regardless of height and perceived importance, must be noted and taken into account
when routing the vehicle. Inconsistencies between the satellite images used to plan

22
missions and the field conditions can result in crashes and improper coverage. While the
satellite images are extremely helpful in planning in a general sense, the images tend to
be outdated and may not be able to fully convey the field conditions in an environment
undergoing rapid erosion. While observing field conditions, it is equally important to note
prevailing winds and any hazards downwind, should the vehicle drift in a strong wind.
Areas where severe damage can occur to the vehicle, property, or people should marked
and all attempts to avoid these areas should be made.
The mission planning software used for this study was Michael Osbornes Mission
Planner version 1.3.10 (Osborne, Mission Planner, 2014). This utility is an open-source
program written in C++. The open-source nature of the program renders the program free
to anyone, with constant updates being pushed to users. These updates incorporate userfeedback and have recently begun incorporating features to allow UAV operators to
avoid restricted airspace and other areas of concern. The open-source nature also
generates a large and responsive community of hobbyists. This resource is incredibly
valuable and makes the Mission Planner stand out among other programs. It has been
adopted by several remote-controlled vehicle manufacturers and auto-pilot developers.
The Mission Planner is also fully capable of running on in a Windows environment as
well as an Android environment. To date, this study has utilized a Dell Latitude D830
laptop, a Dell Inspiron laptop, and an Alienware 14 laptop, all running different versions
of Windows, from XP to Windows 8. The Mission Planner has consistently worked and
has proven to be very reliable. The Mission Planner has also been integrated into a Dell
Venue 7 tablet. With this configuration, the mission planning phase of the methodology

23
can be done with physically in the field, thus allowing for much greater coverage and
efficiency during the flight.
This study used the Mission Planner utility in the following manner. The location to
be surveyed was found on the Google Earth interface within the planning window.

Figure 3.2 Mission Panners Flight Plan window with Google Earth interface
Once the area was located, the home location for the UAV was set by moving
the green Home marker to the approximate location of the takeoff and landing site. It is
important to decide what settings will be used with regards to the Home function, as the
UAV can be programmed to automatically set the point of takeoff as the Home
coordinate, or it can rely on the manually placed marker. In either case, setting the
Home marker within the study area allows the utility to be opened and closed without
having to re-center the map constantly. After the Home location was set, a bounding
box was drawn around the study area, being careful to avoid areas of hazards, as
previously discussed.

24

Figure 3.3 Bounding box around the area to photographed by the UAV
Once the bounding box was constructed, the user can choose to have a survey
pattern generated automatically by the program. The program takes into account user
variables such as camera type, imaging sensor resolution, desired altitude, desired
airspeed, overlap, and side lap. These inputs generate a flight plan that the vehicle then
uses during the mission. These functions are accessed through the Auto-WP menu after
right-clicking on the map.

Figure 3.4 The Auto-WP interface, allowing for automatic generation of survey
waypoints

25
For this study, a Sony NEX-5R was used as the imaging sensor. The lens chosen
was a 16 mm pancake lens. This particular combination was present on the Mission
Planners list of pre-programmed camera options. The program used the information
stored for the particular camera and generates footprints, coverages, and waypoints based
on the user variables. For this study, 80% overlap and 80% side lap inputs were used.
While a 30% side lap and 60-80% overlap is typical in photogrammetry studies, the area
chosen was small and flights using standard parameters were calculated to be less than
one minute in duration (Agisoft, 2014). The coverage was visualized prior to flights
through the footprint function found in the Auto-WP dialogue box.

Figure 3.5 Photo footprints calculated from the Auto-WP function of Mission Planner
While the study used an intervalometer as discussed in the Components section,
the Auto-WP dialogue allows the user to program a camera to trigger based on several
different parameters: distance covered, time elapsed, or manually entered waypoints. This
function would be very helpful for missions involving complex environments or missions
that can only image certain areas along its flight path. The Auto-WP dialogue also
displays values for ground resolution, number of images taken, distance covered, and

26
flight time. These values are very useful in gauging the performance of the flight and
setting limitations to be followed during the flight.
Within the Auto-WP dialogue, the user can also set takeoff and landing
waypoints. The addition of these waypoints makes the flights truly autonomous, with
takeoff and landing being controlled by the autopilot. This feature is useful in areas
where precision landing is not required. This study found that, while excellent for
takeoffs, the landing function is not a fluid as a human pilots landing.
Once the generated mission has been manipulated and checked, the waypoint file
can be uploaded directly to the vehicles autopilot. This is done through the telemetry
link with the computer and can be performed offsite or in the field. Being able to upload
the flight plan in the field is beneficial, as during this study, conditions in the field
required changes to the original flight plan that were not accounted for while planning
offsite. Such conditions include wind, cloud cover, manmade structures, and the presence
of people.
It is important to note the following photogrammetry principles which must be
accounted for during the Mission Planning phase. First, the camera is mounted in a nearvertical orientation. The resulting focal point of the lens is close to the Nadir point.
Typically, photogrammetry projects utilized top-down imagery that would produce
clear orthophotos. The top-down approach minimizes distortion when stitching
photographs together.

27

Figure 3.6 Diagram of Nadir Point (N), Focal or Principle Point (P), and Angle of
Incidence or Tilt. Source:
http://userpages.umbc.edu/~tbenja1/umbc7/santabar/vol1/lec6/image1mh.jpg
For 3D model generation, a portion of the photos used must contain an angle of
incidence of greater than 5. Utilizing too many top-down images results in poor depth
perception for the computer vision algorithms. For the computer to generate correct z
values, or heights, the computer must be able to see the sides of objects and terrain.
Figure 4.5 depicts the results of a flight in which one side of a lifeguard tower was not
photographed, and the computer extrapolated points based on nearby points. The
resulting DEM shows a linear slope from the elevation of the last photographed point to
the next photographed point available.

28

Figure 3.7 Depictions of extrapolated DEM by Agisoft PhotoScan Pro for areas with no
data. Top image is a textured 3D model, bottom image is DEM rendered with no photomosiac
During the course of this study, it was determined that the use of the hard-mount
attachment for the camera resulted in the camera experiencing angles of incidence greater
than 5. This is due to the natural movements of the UAV and the inability of the camera
and mount to compensate due to the fixed-attachment point. It is important to note that if
the camera system utilized a passive gimbal, while it would provide much smoother

29
images, it would not be able to satisfy the angle of incidence requirement for proper 3D
model generation. In order to both use a gimbal system and obtain high quality 3D
models of a scene, the UAV would need to be equipped with a motorized gimbal that
could interface with the APM 2.6 (Robotics, Tarot T-2D Brushless Gimbal Kit, 2014).
These devices do exist, and can be programmed to provide the necessary photographs,
however they are complex and add unneeded weigh to the UAV.
3.2 SITE PREPARATION
In the Site Preparation stage, it was important to take as many measures as
possible to ensure a positive outcome of a flight. During Site Preparation, it was
extremely important to assess light conditions, weather conditions, and physical
conditions of the site. The quality of the resultant digital elevation models depended upon
the steps taken during site preparation to optimize the site under the present conditions. In
certain conditions, a failure to alter flight plans, place GCPs in the optimal location, or
failure to account for environmental variables may lead to insufficient image quality to
produce a usable digital elevation model.
The placement of the Ground Control Points (GCPs) is also a very important
aspect of site preparation. While it is possible build a 3-dimensional model of a site with
photographs and no ground control points, the ground control points are the coordinates
that enable the model to utilize a specified unit, such as metes, or be physically located
within a coordinate system. It is therefore imperative that ground control points are
carefully planned and thoughtfully placed. According to the software developer, Agisoft,
a minimum of 5 GCPs must be used, preferably 10-12 GCPs depending on the site. In
addition, the GCPs must be placed in uniform locations and cover the corners of the
area to be mapped (Agisoft, 2014). The placement of GCPs at the corners of the site to be

30
mapped allows the software to interpolate the positions of points within the site without
significant warping (Agisoft, 2014).
The creation of each GCP should follow certain guidelines. Most importantly, the GCP
needs to be identifiable in the images taken by the UAV. This requires each GCP to be
large enough to be seen by the imaging equipment, in this case a Sony NEX-5R, at the
target altitude of the planned mission. Failure to create large enough GCPs will result in
images with hard-to-identify GCPs that may hinder geo-location and warping.
Conversely, a well-constructed GCP that is significantly large enough for the mission
will be easy to spot, read, and reference during data processing. Furthermore, the
Agisoft PhotoScan Pro software has the ability to recognize GCPs; making the GCP
easily identifiable reduces the errors in the software programs portion of automated GCP
placement. The use must still validate the placement of each GCP in each photo, but by
using clear GCPs that are visible in the photo and good quality images, the software
reduces the time of this usually time consuming step (Agisoft, 2014). In this study, the
GCPs were constructed of plywood. Each GCP measured approximately 30 cm by 30
cm. The top side of the wood was painted flat yellow. It is important to select a nonreflective color such as flat yellow in lieu of gloss. The flatness of the color prevents light
from shining off the GCP and potentially obscuring the GCP in the photo. The yellow
was contrasted by flat black numbering on each GCP. The contrasting colors make the
GCP very easy to read at heights over 37 meters. The numbering on each GCP allowed
for easy identification and made referencing much simpler after the flight was completed.

31

Figure 3.8 Ground Control Point design


In preparing a site for a flight operation, it is necessary to determine how many
GCPs will be needed and how they will each be placed. On a relatively flat surface, such
as a heavily trafficked beach, GCPs can be placed in a simple box pattern, with each GCP
an equal distance from the last. This configuration will allow the program to interpolate
each individual point in between the GCPs without warping. After final placement of the
GCPs, it is necessary to record the location of each GCP. For this study, a Garmin Etrex
GPS was used. While not ideal for survey grade GIS projects, this unit represents the
standard recreational-level GPS unit available. It is accurate to within 3 meters. GPS
units typically have a Z-accuracy of two or three times less than the X,Y accuracy. Due to
this, the Garmin Etrex GPS Z-accuracy is between 6 to 9 meters (Garmin, 2005). This
margin of error prevents the Garmin Etrex GPS from being used to coordinate survey-

32
grade Digital Elevation Models. However, the GPS unit is capable of revealing trends
and providing a coastal manager with a clearer picture of the beach ecosystems dynamic
responses to storm events and natural erosion.
For this study, it was determined that an appropriate number of GCPs for the site
would be between 8 and 12 1. For two flights 12 GCPs were used, while the final flight in
September utilized only 8 GCPs. It is important to note that not all GCPs will be usable in
the final model. Using 8-12 GCPs provides enough confidence that the GCPs captured
will allow for proper orientation and geo-referencing. The most important GCPs to
capture are the GCPs that form the border of the study area. Without these points, the
model may experience warping issues because it cannot determine the full extent of the
model 2. For this study, it was also necessary to place GCPs on the high-high water line,
in order to establish a zero baseline.

1
2

Personal Correspondence with Agisoft Support


Personal Correspondence with Agisoft Support

33

Figure 3.9 Ground Control Points displayed in Agisoft Photoscan for the May 22, 2014
DEM

Figure 3.10 Ground Control Points in aerial photograph for June 25, 2014 DEM

34
Once the GCPs were placed, their positions were recorded. For this, the Garmin
Etrex GPS unit was used. In order to zero the elevations to the high-high water line at
the time, the altimeter was reset at the GCPs that are placed along the line. These GCPs
form the baseline that all other GCP elevations are based on. This method allowed for the
user to only measure the difference in elevation, in lieu of reading an actual elevation
reading which is not accurate, particularly along sea level.
Table 3.0 June 25, 2014 Ground Control Point table with X, Y, and Z values
GCP
#

1
2
3
4
5
6
7
8
9
10
11
12

Elevation
Latitude Longitude (m)
26.441778 -80.060365
1.250
26.441934 -80.060336
1.798
26.442047 -80.060376
1.250
26.442026 -80.060198
1.615
26.441933 -80.060198
0.945
26.441764 -80.060218
1.554
26.441791 -80.060079
1.250
26.441880 -80.060063
2.408
26.442020 -80.060030
1.920
26.442020 -80.059856
0.000
26.441925 -80.059853
0.000
26.441823 -80.059869
0.000

Once the GPS values were recorded, they were entered into Microsoft Excel. For
this study, the data utilized a text format (.txt) that was tab delimited. This format worked
best for importing into Agisoft PhotoScan Pro. The table must be imported as shown
above; improperly imported tables may result in misplacement of the DEM and improper
orientation.

35
3.3 FLIGHT OPERATIONS
The flight operations portion of this study were based on best practices among the
radio-controlled hobby community. Flight procedures and safety measures were designed
to mimic the Academy of Model Aeronautics (AMA) own guidelines for recreational
remote control (R/C) flying (AMA, 2014). This aspect is extremely important, as AMA is
the premier R/C association, with accepted standards and practices acknowledged by the
FAA (Egan, 2014).
Prior to launch of the UAV, it was necessary to do a pre-flight checklist (Osborne,
Pre-Flight Checklist, 2014). As with any manned aircraft, each system must be checked
to ensure proper operation and safety for the flight. The pre-flight check allows the
operator to discover any issues that may render to UAV incapacitated after takeoff. These
failures include battery disconnection, motor failure, propeller loss, and GPS failure.
These failures are easy to detect and remedy prior to launch, but could result in
significant damage to the airframe if not properly dealt with. The checklist was broken
down into several separate sections. First, the UAV was connected to the battery to allow
for all systems to start up. During this time, the GPS was given time to acquire an
adequate signal lock with as many satellites as possible. After connecting the battery the
following order of checks was utilized:
a. UAV component check
i. Check the airframe for signs of weakness,
cracks, or other structural deficiencies
ii. Ensure all bolts and nuts are secured, prior to
check, all nuts should be fastened using a thread

36
locking compound such as Lok-Tite, as this will
prevent vibrations causing the nut to back out
and fall
iii. Check the motors and connections to the arms,
any weakness at the motors will result in motor
failure
b. Electronics check
i. Check all wires and connectors for proper
operation
1. Loose wires may allow a mission to
begin, and then cause a catastrophic
failure in-flight
ii. Radio Link
1. Ensure radio link between UAV and
ground station is established
2. Ground station relays all data to operator
from UAV
iii. GPS calibration
1. Confirm that the GPS has acquired a
signal lock and is receiving a strong
signal

37
2. Ensure the connection with the ground
station is strong and the GPS signal is
accurate
iv. Gyro calibration
1. Allow for gyro-scope to self-calibrate
after battery connection
2. Test gyro performance by manipulating
entire airframe, watching the ground
station to ensure each movement is
accurately depicted
a. Improperly calibrated gyroscopes
will cause crashes of airframes
c. Payload check
a. Secure the camera to the UAV
i. Secondary connections should be utilized to
prevent the camera from falling if the primary
connection fails
ii. Connections should by tight, play will result
in a shifting center of gravity, possibly throwing
off the gyroscope
b. Properly orient the lens of the camera
i. Camera should face the ground

38
1. Nadir point should be as close to focal
point as possible
2. Camera lens should remain in fixed
position
a. Allows for easier calculation of
angle of incidence
ii. Ensure proper script is loaded into camera
1. Intervalometer script allows camera to
take photographs at pre-determined
intervals, removing the need for input
from the operator
iii. Check camera system settings to make sure they
are optimized for ground conditions
1. Shutter speed
2. Exposure
d. Flight Conditions check
a. Make sure all environmental variables are within predetermined safe range
i. 3DRobotics Y6 able to fly in winds 5-10
kph, with camera load attached
b. Ensure the area is clear of any unnecessary people
Following the pre-flight checklist, it was imperative that the UAV was monitored
at all times. Additionally, all non-essential people were cleared of the study area. Once

39
the pre-flight checklist had been completed and all problems dealt with, it was possible to
proceed with loading the flight plan into the UAVs autopilot. The flight plan contained
several safety features that were checked and adjusted to match the conditions in the
field. These features are a series of rules that, if broken, will cause the UAV to return to
its point of takeoff. Geo-fencing is a rule in which the UAV is prohibited from leaving an
area defined by coordinate points which encircle the target area. If the UAV breaks this
fence, the autopilot will disable the mission and return to the point of takeoff. This
feature is one of the most important safety features for UAVs operating in densely
populated or built-up areas. During this study, the UAV was only permitted to fly above
the citys park, and not over private areas or roads. To ensure those conditions were met,
a geo-fence was programmed into every mission that encompassed the small study area.
In addition to geo-fencing, the 3DRobotics Y6 UAV is capable of setting a maximum
altitude. This feature, like geo-fencing, is an important public safety feature. As a
recreational model, the UAV was prohibited from flying above 400 feet AGL. To meet
this criteria, the altitude maximum was set at 350 feet AGL, however, no mission was
programmed higher than 150 feet AGL. If control of the UAV was lost, or if the payload
falls off and the UAV gains significant altitude without user input, the altitude maximum
rule will override the auto-pilot or operator and return the UAV to an allowed altitude.
Flying the UAV in auto-pilot mode with a pre-programmed mission creates
several advantages for the operator. With the autopilot commanding the UAV, the
operator was free to observe the UAV without having to focus on controls. While the
controls need to be in the operators hand while the UAV is in the air, the freedom to
observe without manipulating the controls allows the operator to notice problems quicker.

40
The operator is able to focus on the UAV for longer periods of time rather than
monitoring the controls frequently.

Figure 3.11 Mission Planners flight screen displaying telemetry data and gps
information

41

3.4 DATA PROCESSING

Figure 3.12 Conceptual map of Data Processing phase


During the Data Processing stage of the project, the images were loaded into
Agisoft PhotoScan Pro and processed using Structure from Motion technology. In
this stage the operator generated both the point cloud and the digital elevation model. It
was important that the user carefully analyzed the images available, and selected only the
best quality images. The Structure from Motion software attempts to place a point in a
three dimensional box by comparing the points location in at least three separate images;
the more images available, the more chance there is for the software to incorrectly place

42
the point. The study found that the Photoscan software produced much cleaner models
when only the best images (roughly 20-40% of the images captured) are used than when
all of the images are used. This is due to several factors noticeable on the images
themselves. First, if an image is blurry, the software will not be able to place the point
accurately, resulting in skewed models. Second, shadows can significantly hinder the
quality of model. While an image may be of good quality, the program has issues when
heavy shadows are present. Third, if the point is visible in many images, the software
program may not identify that point in each image, and thus assign a new point. This will
impact the quality and accuracy of the final model. To ensure proper operation of the
Structure from Motion software, Agisoft PhotoScan Pro, the following procedure
was used:
1. Select the best photos in the dataset
a. Photos should have:
i. Good lighting
ii. Good resolution
iii. Contain at least 1 Ground Control Point
iv. Differ from previous images
2. Begin a new Photoscan project
3. Load selected photos into the project by utilizing the Add Photos
button
4. Once photos have been loaded, load in Ground Control Point data in
.txt file

43
a. .csv extentions dont work as smoothly between programs as
.txt, tab delimited files do
5. Review the information presented by Photoscan regarding the GCPs
a. GCPs without Z-values will be omitted entirely
6. Place GCPs in each photo
a. Select the photo from the photo browser pane
b. Right click on the GCP
c. Select Place Marker
d. Select the corresponding marker
e. Repeat this until all GCPs in every photo are mapped
7. Verify the GCP accuracy
8. Assign a projection system to the project
a. Select the Ground Control Preferences icon
b. This project used the WGS 1984 ellipsoid
i. GPS unit referenced same ellipsoid
ii. Ensure the GPS and Photoscan are using the same
system
9. Begin model generation by selecting Align Photos
a. Select High Quality, Generic options in settings tab
10. Optimize the alignment of the sparse point cloud
a. Best results were found to be when all except align k4 were
selected
11. Run Build Dense Point Cloud

44
a. Medium quality is preferred
i. High quality provides too much detail for a reliable
DEM, increases processing time and required resources
ii. Low quality may not provide enough accuracy
12. Run Build Mesh
a. Select the Surface Type: Arbitrary
b. Select Source Data: Dense Cloud
c. Select Polygon Count: Medium
i. This will prevent too many faces being generated
13. Run Build Texture
a. This will overlay the orthphoto over the DEM, giving an easy
reference map for DEM geo-locating in other programs
14. Select Export Orthophoto
a. This will export an orthophoto of the study area
15. Create a Digital Elevation Model
a. This will export a DEM to a specified folder
b. Use the (.tif) for best results
16. Close Photoscan Pro
17. Open ArcGIS for Desktop
18. Load the DEM into ArcGIS using the Add Data tool
19. Ensure the DEM is utilizing the correct geographic projection system
a. This should match the geographic projection system selected in
PhotoScan

45
20. After reviewing the DEMs properties, alter the appearance of the
DEM by changing the color bar and display symbology to the most
appropriate option
a. This study utilized a green-to-red color bar to display areas of
high and low elevations
21. Repeat the steps for a DEM generated for another time frame
a. The second DEM should cover the same area as the first
i. Any areas not overlapping will be excluded from
analysis
22. Once both DEMs are loaded into ArcGIS, there are two options
a. Using the Spatial Analysis Map Algebra Tool, subtract the
older DEM from the Newer DEM
i. If using this method, multiply the resulting DEM by -1
1. This will produce negative values for erosion
and positive values for accretion
b. Using the Spatial Anaylsis Map Algebra Tool, subtract the
newer DEM from the older DEM
i. This method eliminates the need to multiply by -1
23. Display the resulting DEM in a matter most appropriate for the project
a. In this case, the DEM was displayed using a Red (erosion) to
Green (accretion) color scale
24. The resulting DEM allows for visualization and evaluation of the area

46
After the DEM was created, the data was analyzed using the Profile Line tool in
ArcGISs 3D Analyst toolbar. This tool, along with others in the toolbar, allows the
user to draw a line through a DEM. The program can then analyze the DEM and generate
a topographic profile of the DEM, from the starting point to the end point. This tool is
very useful in beach erosion monitoring because the beach profiles most commonly used
to monitor and quantify erosion are created in the same manner.

CHAPTER 4: RESULTS
The UAV performed 9 days of flights, gathering approximately 7 gigabytes of
aerial imagery in that time frame. Due to environmental conditions, flight issues, and
component issues, only three days of flights produced usable data for Digital Elevation
Model generation. While every day produced enough data to accurately model the beach
ecosystem in 3 dimensions, only the three flights discussed further utilized Ground
Control Points that allow for proper geo-referencing. Of the days that produced usable
results, the average flight time was 4 minutes and 52 seconds. In May and June, two
flights per mission were utilized; while in September, only one flight was flown.
The Structure from Motion software was able to accurately capture the spatial
relationships between identifiable objects within the scene. For each flight, the
photographs taken were able to generate plausible 3D models. However, without the
Ground Control Point verification, these models cannot be transferred into a coordinate
system, rendering them unusable. For the three flights that did utilize proper Ground
Control Points, the 3D models easily transferred into Digital Elevation Models for use in
ArcGIS.
Flight results were positive. The UAV was able to complete mission plans
without issue. For safety and equipment concerns, 50% capacity was left in the batteries.
When the battery capacity falls below 50%, the power of the UAV tends to decrease
dramatically. Often times, this result in an overcorrection as the UAV descends rapidly.
Additionally, flights were kept at low altitude. This resulted in higher resolution and
more images. The additional images resulted in longer processing times. One benefit to
the large number of images generated was the ability to discard images with flaws. If the
47

48
UAV was flown at higher altitude, it would not need as many images to cover the entire
study area, which may result in poor results if some images prove to be unusable. Factors
affecting the quality of the images during these flights include vibrations in the images,
glare, shadows, and out-of-focus images. It is impossible to compensate for shadow and
glare when developing a flight plan, and a larger number of images were required in
order to minimize the impact of these factors. To deal with out-of-focus images, the iso
settings were placed on auto, allowing for auto-focus. The fixed 16 mm lens also
simplified the focusing, resulting in fewer out-of-focus images. The Sony NEX-5R has a
relatively fast image sensor and can focus in a very short period of time. To mitigate the
vibrations, the shutter speed was set to the fastest setting on the camera.
The data collected revealed several interesting trends. First, the DEMs revealed
that the beach experiences both erosion and accretion at the same time. The landward
portion of the beach tended experience erosion, the middle portion of the beach tended to
experience slight accretion, and the portion closest to the tidal line experienced the most
erosion. It is important to note that the DEMs extend past the area within the ground
control points, resulting in inaccurate elevation values. These values are not relevant for
this study because the study focuses on the tidal zone, and areas with inaccurate values
reside within the vegetation area of the beach.

49

Figure 4.0 Profile Lines representing the slope of the study area along a set of profile
lines. The large spike represents the lifeguard tower

Figure 4.1 Profile Lines representing the slope of the study area along a set of profile
lines

50

Figure 4.2 Profile Lines representing the slope of the study area along a set of profile
lines

51

Figure 4.3 Elevation changes between September 2014 and May 2014. Red indicates
erosion, while green indicates accretion.

52

Figure 4.4 May 22, 2014 DEM

53

Figure 4.5 June 25, 2014 DEM

54

Figure 4.6 September 19, 2014 DEM

55

Figure 4.7 May 22, 2014 Ground Control Point Data

56

Figure 4.8 June 25, 2014 Ground Control Point Data

57

Figure 4.9 September 19, 2014 Ground Control Point Data

58

Figure 4.10 May 22, 2014 Point Coverage Map

59

Figure 4.11 June 25, 2014 Point Coverage Map

60

Figure 4.12 September 19, 2014 Point Coverage Map

CHAPTER 5: DISCUSSION
Using an Unmanned Aerial Vehicle to monitor coastal erosion is both feasible and
economical. As this study shows, with proper training and techniques, it is possible to
obtain high-quality, high-resolution aerial photographs of a coastal environment. The
methods described above can be used to produce Digital Elevation Models, Orthophotos,
and analytical models that can be used to track coastal erosion. The combination of both
the Digital Elevation Models and the Orthophotos enables the user to utilize more than
one method of erosion monitoring. This aspect enhances the probability of a good
analysis, even if one particular method does not work out. With the low initial cost of the
unmanned aerial vehicle, software, and training, this method provides small
municipalities and coastal managers the opportunity to map and monitor coastal areas
that are too small for standard LiDAR projects. Typical LiDAR projects are priced per
square mile, with minimum area requirements. Small LiDAR projects have been
estimated to cost roughly $750- $1000 per square mile (Hallum & Parent, 2009).
Some small towns, such as Delray Beach, only have 2 to 3 miles of coastline. The
area to be mapped would not cover the operating costs of an airplane with the LiDAR
module, as well as the cost to process the information be a trained LiDAR technician.
Structure from Motion technology such as PhotoScan allows for the use of cheap
digital cameras and moderately-equipped computers to achieve similar results to LiDAR.
The range limitations of UAVs combined with the cost restrictions of LiDAR make small
scale projects a prime candidate for the proposed method. In addition to small areas,
Structure from Motion techniques and UAVs can be used as quick-response tools
following significant weather events. A hurricane or large swell event will cause drastic

61

62
erosion that is not typically mapped. With a cheap, easy to use system available, a
municipality can track how each storm impacts the coastline. This will provide better
information to the municipality regarding the effectiveness of beach nourishment and the
impact of single events on coastlines.
For this study, the Digital Elevation Models produced by the Structure from
Motion software were suitable for tracking elevation changes caused by erosion. The
models were able to mimic the real-world observed conditions and captured the spatial
relationships with sufficient detail. With the use of ArcGIS as the main analytical
software for the Digital Elevation Models, the process of analyzing the DEMs is very
similar to the process used to analyze LiDAR data. In addition, the 3D models generated
by PhotoScan can be saved and exported as .las, a LiDAR point cloud dataset. Other file
extensions available are Wavefront (.obj), Stanford PLY (.ply), XYZ Point Cloud (.txt),
and U3D (.u3d) 3. The multitude of file types available during model export in PhotoScan
allows the models to be used in a variety of software suites including ESRI ArcGIS and
Autodesk software such as AutoCAD and Map3D.
This methodology can be used for any Digital Elevation Modeling project, not
just coastal erosion monitoring. The main components in the methodology are the GPS
unit, the imaging sensor, and the computer. The major limitation experienced during this
study was the accuracy of the GPS unit. The Garmin Etrex unit used did not have an
antenna capable of receiving the L1 or L2 GNSS signals, resulting in accuracies of 2-3
meters. If this methodology is paired with RTK GPS units, the accuracy would rival
LiDAR surveys.
3

Agisoft PhotoScan Pro Interface

63
When using an Unmanned Aerial Vehicle, it is important to take into account
basic photogrammetric principles. In order for the Structure from Motion software to
produce accurate 3D models, it is necessary for some of the photos to be taken at 5 angle
to nadir point. Fortunately, with the fixed camera mount and the natural motion of the
UAV, it is possible to obtain 5 angle photos without altering flight plans.

Figure 5.0 PhotoScan project depicting camera locations and angles (blue squares with
vectors) relative to the DEM. 5 angle from the Nadir point is possible and usually done
in normal course of flight.
The use of unmanned aerial vehicles, or drones, in recent years has drastically
increased. With the development of truly open-source autopilot systems and cheap-toproduce parts, unmanned aerial vehicles have dropped in price to within reasonable reach
of ordinary consumers. With this drastic decrease in price and the relative ease of flying
with a computer-assist, the popularity of drones has exploded. While unmanned aerial
vehicles have been around since the 1950s, they have not existed outside the realm of

64
aviation-minded hobbyists and the military until the latter half of the last decade. In
conjunction with the greater availability of drones, camera systems have also experienced
dramatic technological improvements and cost reductions. These factors have created a
very popular new trend in which video and photographs are captured from drones. These
drones are being flown in almost any area they can be. With the popularity of these
drones and the activities they are used for, there has been a serious concern over privacy.
The general consensus of the public is that the ease at which these systems are operated
and their low cost make privacy invasions much easier for criminals. While these
concerns are legitimate, the benefits presented by this new technology should not be
overshadowed by possible misuses.
In addition to privacy concerns, safety is a high priority for the FAA. With the
low cost and high performance of new UAVs, operators can achieve altitudes and ranges
that previously wouldve taken a skilled pilot to reach. This creates the opportunity for
adverse manned flight and unmanned vehicle interactions, particularly around airports.
While the FAA has yet to release any rules pertaining to UAV operations (they have
issued several internal memos), the model airplane community is pushing for safety
guidelines modeled after the AMAs set of rules. These rules keep operators away from
populated areas, flying with supervision, and limit the altitudes of the vehicles. While it is
not desirable to fully adopt the AMA rules due to their requirement to fly within a
designated R/C airfield, they have an outstanding safety record and members of the AMA
hold themselves to a high standard. Along with the AMA, several UAV hardware and
software developers have begun releasing free safety updates. These updates include
changes to the GPS and map interfaces in mission planning software that display no-fly

65
areas around airports and other sensitive sites. These sorts of developments help keep
UAV operators safe and prevent any unwanted interactions with the public or FAA. More
safety protocols need to be developed for UAVs, but at this early stage in UAV
technology, several developers have already made significant progress.
5.1 COST
The method proposed above has many advantages over standard LiDAR surveys.
It can be done more frequently, more irregularly in response to specific events, and can
be done with significantly less planning involved. Perhaps the biggest difference between
LiDAR and Structure from Motion technology is the cost. The program used for this
study, PhotoScan Pro, retails for $3000, on par with other computer programs of similar
type. However, for educational institutions and researchers, the program can be
purchased for $500. The Y6 unmanned aerial vehicle cost roughly $1200. The laptop
was mildly built for advanced computing and cost $1500. This cost should not be
factored into a cost analysis for this method because the computer hardware required
between LiDAR and Structure from Motion techniques is similar. The camera chosen,
the Sony NEX-5R, cost $400 with an additional $175 for the lens. In all, including the
laptop, the total investment for this project was less than $4,000.

66
Table 5.0 Total investment in system.
Item
Agisoft
PhotoScan Pro
Alienware 14
Laptop

Upgraded Battery
16 mm Pancake
Lens

Cost
$
500.00
$
1,500.00
$
1,200.00
$
400.00
$
50.00
$
175.00

Total Investment

$
3,825.00

3DRobotics Y6
Sony NEX 5R

5.2 FUTURE RESEARCH


As this study demonstrates, Unmanned Aerial Vehicles are a useful tool now
available to coastal managers. When used in conjunction with Structure from Motion
software, UAVs can provide coastal managers with a unique perspective of the
ecosystem, allowing for better management and more frequent monitoring. However, this
technology is still in its nascent stage, and several areas can be significantly improved.
Most notable, the onboard GPS can benefit from RTK technology. This technology is
still in early development, however products such as Swift Navigations Piksi module are
available and easily integrate with the APM 2.6 (Swift, 2014). UAVs and this
methodology could benefit from RTK GPS sensors through more precise navigation and
situational awareness, formation flying, and more exact camera locations resulting in
faster post-processing times. Further research should also investigate the use of airplanestyle UAVs for larger survey areas. While this method is theoretically scalable, as long as

67
the four corners are referenced and a predetermined amount of Ground Control Points are
used, the logistics of larger areas combined with using an airplane-style UAV could vary
significantly from this multi-rotor based methodology.
5.3 APPLICATIONS
The methodology proposed here has further uses beyond coastal erosion
monitoring. The workflow can be used to map large areas of forest and develop tree
height maps. This would be accomplished through further steps within Agisofts
PhotoScan Pro which would classify points into groups and allow the user to define
ground and tree groups. The resulting Digital Terrain Model would illustrate tree height
above the ground. Another potential use for this method would be in tracking tidal
changes over time. By building both Digital Elevation Models and Orthophotos, the
extent of the tidal inundation of an area could be accurately tracked over a period of time.
This would enable a clearer understanding of how sea level rise will impact a specific
area. Other potential uses for this methodology include fuel-load estimates for wildfire
management, flood risk analysis using DEMs and ArcGIS, archaeological site
documentation, disaster and storm response documentation, and rangeland and coastal
restoration.

CHAPTER 6: CONCLUSIONS
Structure from Motion technology, coupled with Unmanned Aerial Vehicles,
can provide valuable information to a coastal manager that may not be available p
reviously. Due to the low cost nature of this method, surveys can be performed on a
regular basis with minimal planning and cost. In addition, since this method shifts the
majority of the work from the field into an office, costs are further lowered. As shown,
using this method can result in the generation of a usable Digital Elevation Model in 12
to 14 hours, for this study area. Of those hours, only 1 hour is dedicated to field work.
The remaining hours are not dependent on weather. Further, the time can be reduced by
utilizing higher-end computer systems. This study was conducted on a relatively
inexpensive laptop, and the skills required to complete this method are comparable to
most GIS skills, particularly operation of field equipment. This means that this method
would be fairly simple for a GIS technician to adopt and utilize. The results generated by
this technique can be comparable to LiDAR data, assuming the proper gear is utilized
(Eltner, Mulsow, & Maas, 2013).
While the 3D modelling portion of the technique is very accurate, it relies on
accurate ground control points for realistic Digital Elevation Model generation. This
aspect was a major hindrance to this study. While the data acquired was able to detect
trends and provide graphic representation, the values were not precise enough to be
compared or used for survey-grade applications. Consumer-level GPS devices are simply
not accurate enough, particularly on the Z-axis, for this type of work. An appropriate
replacement technology would be Real-Time Kinematic GPS systems that utilize a base
station and a rover. This technology provides the accuracy of LiDAR and can be

68

69
integrated in the UAV. Devices such as Swift Navigations Piksi RTK receiver allow the
ground station computer to be used as the base station, and the UAV is utilized as the
rover. Other traditional RTK and GNSS systems are also applicable for this methodology.
Further investigation is required in the area of Z-value accuracy and GPS integration with
Structure from Motion software and UAV flight operations.

Works Cited
Agisoft. (2014). Orthophoto and DEM Generation (with GCP) with Agisoft PhotoScan
Professional. Retrieved from Agisoft PhotoScan:
http://www.agisoft.ru/wiki/Photogrammetry
AMA. (2014, January 1). Academy of Model Aeronautics National Model Aircraft Safety
Code. Retrieved from Model Aircraft:
https://www.modelaircraft.org/files/105.pdf
Coleman, C. P. (1997). A Survey of Theoretical and Experimental Coaxial Rotor
Aerodynamic Research. NASA Technical Paper 3675.
Drones, C. (2014). The Mission of ConservationDrones.org. Retrieved from
Conservation Drones: http://conservationdrones.org/mission/
Egan, P. (2014, January 11). MOU between the FAA and AMA. Retrieved from sUAS
New: http://www.suasnews.com/2014/01/26919/moa-between-the-faa-and-ama/
Eisenbeiss, H., & Zhang, L. (2006). Comparison of DSMs Generated From Mini UAV
Imagery and Terrestrial Laser Scanner in a Cultural Heritage Application. ISPRS
Commission V Symposium 'Image Engineering and Vision Metrology', 90-96.
Eltner, A., Mulsow, C., & Maas, H.-G. (2013). Quantitative Measurement of Soil Erosion
From TLS and UAV Data. International Archives of the Photogrammetry, Remote
Sensing, and Spatial Information Sciences, 119-124.
FLDEP. (2013, February 28). Why Restore Eroded Beaches? Retrieved from Florida
Department of Environmental Protection:
http://www.dep.state.fl.us/beaches/programs/becp/restore.htm
Garmin. (2005). eTrex Vista. Retrieved from Garmin:
http://web.gps.caltech.edu/gislab/Equipment/eTrexVista_OwnersManual.pdf
Hallum, D., & Parent, S. (2009). Developing a Business Case for Statewide Light
Detection and Ranging Data Collection. Retrieved from UNL- Nebraska Water
Center: http://watercenter.unl.edu/PRS/PRS2009/Posters/Hallum%20Doug.pdf
Houston, J. (2013). The Value of Florida Beaches. Shore and Beach.
Little, D. (2014). Beach Renourishment Project. Retrieved from City of Delray Beach:
http://mydelraybeach.com/planning-and-zoning/coastal-projects/beachnourishment-project
Lucieer, A., Robinson, S., Turner, D., Harwin, S., & Kelcey, J. (2012). Using a microUAV for ultra-high resolution mutli-sensor observations of Antarctic moss beds.
Remote Sensing and Spatial Information Sciences, 429-433.
70

71
Mancini, F., Dubbini, M., Gattelli, M., Stecchi, F., Fabbri, S., & Gabbianelli, G. (2013).
Using Unmanned Aerial Vehiclels (UAV) for High-Resolution Reconstruction of
Topography: The Structure from Motion Approach on Coastal Environments.
Remote Sensing, 6880-6898.
Osborne, M. (2014). Mission Planner. Retrieved from ArduPilot:
http://planner.ardupilot.com/
Osborne, M. (2014). Pre-Flight Checklist. Retrieved from ArduPilot:
http://copter.ardupilot.com/wiki/checklist/
Paskulin, A. (2013, October). 3D Robotics Drone Helps Farmers with Precision
Agriculture. Retrieved from 3DR Blog: http://3drobotics.com/2013/10/3drobotics-drone-helps-farmers-with-precision-agriculture/
Robotics, 3. (2014). 3DR RTF Y6. Retrieved from 3D Robotics:
https://store.3drobotics.com/products/3dr-rtf-y6-2014
Robotics, 3. (2014). Tarot T-2D Brushless Gimbal Kit. Retrieved from 3D Robotics:
https://store.3drobotics.com/products/tarot-t-2d-brushless-gimbal-kit
Ruppert, T. K. (2008). Eroding Long-Term Prospects for Floridas Beaches: Floridas
Coastal Construction Control Line . Sea Grant Law and Policy Journal, 65-98.
Sony. (2014, September). Time-Lapse. Retrieved from Play Memories Camera Apps:
https://www.playmemoriescameraapps.com/portal/usbdetail.php?eid=IS9104NPIA09014_00-000003
Survey, F. G. (2014, October). County Geological Maps. Retrieved from Florida
Department of Environmental Protection:
http://publicfiles.dep.state.fl.us/FGS/WEB/geomap2/
Swift. (2014). Piksi. Retrieved from Swift Navigation: http://swiftnav.com/piksi.html
Turner, D., Lucieer, A., & Watson, C. (2012). An automated technique for generating
geo-rectified mosaics from ultra-high resolution unmanned aerial vehicle (UAV)
imagery, based on structure from motion (SfM) point clouds. Remote Sensing,
1392-1410.

You might also like