<%BANNER%>

Design and Validation of Autonomous Rapid Mapping System Using Small UAV

Permanent Link: http://ufdc.ufl.edu/UFE0022743/00001

Material Information

Title: Design and Validation of Autonomous Rapid Mapping System Using Small UAV
Physical Description: 1 online resource (115 p.)
Language: english
Creator: Bowman, William
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: georeferencing, mapping, uas, uav
Mechanical and Aerospace Engineering -- Dissertations, Academic -- UF
Genre: Mechanical Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Advancements in inertial measurement, global positioning, and digital photography have allowed the development of small autonomous unmanned aerial vehicles (UAV) to be used for low altitude airborne imaging. The University of Florida has long been interested in using UAVs for wildlife and natural resource management. Digital, high resolution images taken from the UAV can be arranged to create a map of the flight coverage area. These maps can express a multitude of various environmental constituents that can be useful to biologists and ecologists alike. This thesis presents a low-cost, end-to-end system for rapid single image geo-registration of airborne imagery collected from a small autonomous unmanned aerial vehicle.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by William Bowman.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Ifju, Peter.
Local: Co-adviser: Mohamed, Ahmed Hassan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022743:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022743/00001

Material Information

Title: Design and Validation of Autonomous Rapid Mapping System Using Small UAV
Physical Description: 1 online resource (115 p.)
Language: english
Creator: Bowman, William
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: georeferencing, mapping, uas, uav
Mechanical and Aerospace Engineering -- Dissertations, Academic -- UF
Genre: Mechanical Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Advancements in inertial measurement, global positioning, and digital photography have allowed the development of small autonomous unmanned aerial vehicles (UAV) to be used for low altitude airborne imaging. The University of Florida has long been interested in using UAVs for wildlife and natural resource management. Digital, high resolution images taken from the UAV can be arranged to create a map of the flight coverage area. These maps can express a multitude of various environmental constituents that can be useful to biologists and ecologists alike. This thesis presents a low-cost, end-to-end system for rapid single image geo-registration of airborne imagery collected from a small autonomous unmanned aerial vehicle.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by William Bowman.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: Ifju, Peter.
Local: Co-adviser: Mohamed, Ahmed Hassan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022743:00001


This item has the following downloads:


Full Text

PAGE 1

DESIGN AND VALIDATION OF AUTONOMOUS RAPID MAPPING SYSTEM USING SMALL UAV By WILLIAM SCOTT BOWMAN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2008 1

PAGE 2

2008 William Scott Bowman 2

PAGE 3

To Mom, Dad, Michael and Kelley. 3

PAGE 4

ACKNOWLEDGMENTS First and foremost, I would like to thank my immediate and extended family for their unwavering support throughout my a cademic efforts. I would lik e to thank Grandma Thoenes for her gracious generosity and involvement in my life; I wish I could tell Grandpa Thoenes how influential his love for engineering has been on me. It too has become my passion. I would like to thank Grandpa Bowman for providing me with my obvious entreprene urial spirit and strong work ethic. I extend my greatest appreciation to the Lee family for being so supportive of me. I look forward to joining your family. I would like to thank Dr. Peter Ifju for providing me with the means to grow academically. I have always heeded your advice and have a ppreciated your approachable personality. I thank Dr. Franklin Percival for being such a great ment or and friend. I have enjoyed the collaboration among sciences; we have achieved something great. I thank Dr. Mohamed for providing his expertise in Geomatics and Inertial Navigation; it has inspired my work here. Additionally, I thank Kyuho Lee for being a great leader in th e MAV Lab. I have enjoyed learning from you. I thank my friends for being there when I need ed them. I especially thank Adam and Mike for their help and humorous nature on this projec t. You guys have been great to work with. I will miss junk-works. I would also like to th ank Baron for being such a great colleague and friend through grad school. Thanks again to my friends, Larry Taylor and John Lane of the C.O.E. for their support. Thanks again to Kell ey for all her editing and word processing skills! I would like to give my goodbyes to the Univers ity of Florida. I was given the absolute best college experience. There is no other colleg e in the country that can provide all there is to offer here; stellar academics and thre e national championships. GO GATORS!! Finally, I would like to thank Rusty, the tabby cat, for providing me with comforting companionship during the best and worst times in the writing of this thesis. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES.........................................................................................................................8 LIST OF ABBREVIATIONS........................................................................................................11 ABSTRACT...................................................................................................................................13 CHAPTER 1 INTRODUCTION................................................................................................................. .15 Motivation...............................................................................................................................15 The Polaris UAV Overview....................................................................................................17 Georeferencing and Rapid Mapping.......................................................................................18 2 LITERATURE REVIEW.......................................................................................................20 Small Unmanned Aerial Vehicles..........................................................................................20 Aerial Imaging and Georeferencing.......................................................................................21 Digital Mapping......................................................................................................................23 3 POLARIS UAV AIRFAME...................................................................................................25 Construction................................................................................................................... .........25 Assembly................................................................................................................................27 4 POLARIS UAV ELECTRONICS..........................................................................................34 Motor Selection and Controller..............................................................................................34 Actuators...................................................................................................................... ...........35 Autopilot.................................................................................................................................35 Global Positioning System Receiver......................................................................................36 Subsystem Control Hardware.................................................................................................37 Data Storage............................................................................................................................39 Cameras..................................................................................................................................39 Polaris Power Management....................................................................................................41 Ground Control Station......................................................................................................... ..43 5

PAGE 6

5 POLARIS UAV SOFTWARE...............................................................................................51 Firmware....................................................................................................................... ..........51 Graphical User Interfaces.......................................................................................................52 Camera Synchronization.........................................................................................................53 6 CAMERA CALIBRATION AND INITIALIZATION..........................................................58 Calibration Software........................................................................................................... ....58 Initialization Jig......................................................................................................................62 7 RAPID MAPPING................................................................................................................ .68 Coordinate Frames.............................................................................................................. ....68 Direct Georeferencing............................................................................................................71 Direct Linear Transformation Method....................................................................................74 Photo Geo-Registration......................................................................................................... .76 PolarisView Graphical User Interface....................................................................................77 9 MAPPING OUTPUT AND ANALYSIS...............................................................................84 Data Set...................................................................................................................................84 Results.....................................................................................................................................85 10 SUMMARY, CONCLUSTION AND RECO MMENDATIONS FOR FUTURE WORK..100 Summary...............................................................................................................................100 Conclusion............................................................................................................................100 Recommendations for Future Work.....................................................................................101 Georeferencing..............................................................................................................102 Small Unmanned Aerial Vehicle Control......................................................................104 APPENDIX A POLARIS VIEW GENERATED DATASET......................................................................106 B DEVICE PARAMETERS....................................................................................................110 LIST OF REFERENCES.............................................................................................................111 BIOGRAPHICAL SKETCH.......................................................................................................115 6

PAGE 7

7 LIST OF TABLES Table page 3-1 Wing and stabilizer design parameters..............................................................................30 4-1 Root-mean-square errors in the Kest rel v2.2 INS/GPS in near level conditions...............45 7-1 The WGS84 reference ellipsoid values..............................................................................79 8-1 Indicates calculated boresight misalignments for the data set...........................................88 A-1 Compensated_Pic_Data file used for rect ification, area assess ment and mapping in the results section............................................................................................................ .107 A-2 Pertinent data entries for a given target. This data was used throughout the thesis.......108 B-1 Autopilot Sensor Parameters............................................................................................110 B-2 Canon A650 IS Parameters..............................................................................................110

PAGE 8

LIST OF FIGURES Figure page 3-1 Unpainted Kevlar Fuselage used for the Polaris UAV....................................................30 3-2 Polaris wing, horizontal and vert ical stabilizer plan forms..............................................30 3-3 Tensioning acetate film over fiberglass skinni ng creating a smooth external surface......31 3-4 Process for compressing fiberglass curing epoxy to EPS foam wing cores......................31 3-5 Saddle hatch for programming access of PDIB processor and wing connectivity............32 3-6 Horizontal stabilizer tail boom mounting hardware..........................................................32 3-7 Completed Polaris UAV airframe A) Birds eye view B) side view.................................32 3-8 Airboat launch and recovery setup....................................................................................33 3-9 Polaris day and low-li ght flying configuration..................................................................33 4-1 Procerus Autopilot Kestrel v2.2....................................................................................45 4-2 One Watt Aerocom modem is mounted in tail to reduce EMI..........................................45 4-3 Carbon fiber box for autopilot with wate rproof interfacing electrical connectors............46 4-4 Elevator control loop...................................................................................................... ....46 4-5 Throttle control loop...................................................................................................... ....46 4-6 Aileron/Rudder control loop..............................................................................................47 4-7 Furuno GH81 GPS receiver on Polaris UAV....................................................................47 4-8 Custom designed Power Dist ribution and Integration Board.........................................47 4-9 DosonChip SD Memory Card hardware us ed to store synchron ized UAV orientation and location data................................................................................................................48 4-10 KX-141 CCD camera and installation location.................................................................48 4-11 One watt video transmitter (no ante nna) and video control circuitry................................48 4-12 Canon A650 Digital Camera payload................................................................................49 4-13 Unpopulated camera control board....................................................................................49 8

PAGE 9

4-14 Custom designed power module........................................................................................49 4-15 Front and back of Ground Control Station (GCS).............................................................50 5-1 Atmega 128 MCU firmware flowchart..............................................................................56 5-2 Virtual Cockpit software screen shot.................................................................................56 5-3 PolarisLink software....................................................................................................... ...57 5-4 Camera control window.....................................................................................................57 6-1 Images used for camera calibration...................................................................................64 6-2 Internal grid intersection points are base d on square sizes and initial lens distortion coefficient kc......................................................................................................................64 6-3 Lens distortion model of Canon A650. A)Tangential lens distortion. B) Radial lens distortion..................................................................................................................... .......65 6-4 Complete lens distortion model.........................................................................................65 6-5 Comparison of distorted pi cture based on lens parameters. A) Original distorted picture B) Undistorted picture...........................................................................................66 6-6 Exterior camera paramete rs for indoor calibration............................................................66 6-7 Field initializatoin process for dete rmining camera boresight misalignment....................67 7-1 Definition of coordinate frames used for direct georeferening..........................................79 7-2 Definition of Image Frame Coordinate System.................................................................80 7-3 Example of DLT projections from came ra frame origin through image plane to ground coordinates in mapping frame sa tisfying collinearity conditions..........................80 7-4 Digitally Geo-registered Photos.........................................................................................81 7-5 PolarisView GUI software package...................................................................................81 7-6 Offset menu in PolarisView for determin ing camera relative position and orientation....82 7-7 Example of calibration image of taken just before survey flight.......................................82 7-8 Example of area coverage assessm ent generated from PolarisView.................................83 8-1 Ground control targets used for spatial comparison in the mapping process ...................88 8-2 Polaris UAV flight route vers us numeric target locations.................................................88 9

PAGE 10

8-3 Rectified images for data set are rectified and registered from PolarisView into Google Earths interface..................................................................................................89 8-4 All pre-surveyed points were lo cated in registered photos................................................89 8-5 Pre-survey location of the target versus measured values for those targets through the photos in Google Earth......................................................................................................90 8-6 UAV and camera attitudes at the time of commanded camera exposure..........................93 8-7 Residuals for all registered photos.....................................................................................94 8-8 RMS of Euclidean distances for all targets........................................................................97 8-9 Averaged combined pitch and roll eff ects on positional accuracy apparent in data..........97 8-10 Simulated principal point error on the ground from IMU inaccuracies in pitch and roll..98 8-11 Rates of all UAV states dur ing initial 20 phot ograph capture...........................................98 8-12 Unaccounted for effects of an 87 millisecond synchronization error with high dynamics present............................................................................................................... .99 10

PAGE 11

LIST OF ABBREVIATIONS AC/DC alternating current/direct current ACK/NACK acknowledge/no acknowledge ADC analog-to-digital ASCII American Standard Code for Information Interchange CAD computer-aided design CCD charge-coupled device COTS commercial off-the-shelf DARPA Defense Advanced Research Projects Agency EMI electromagnetic interference EPS expanded polystyrene foam GCP ground control point GCS ground control station GIS graphic information system GPS/INS global positioning syst em/inertial navigation system GUI graphical user interface I/O input/output IMU inertial measurement units IP internet protocol KML keyhole markup language KV rotations per minute per volt LED light emitting diode LPP leadless plastic package mAH milliampere-hour MAV micro air vehicle 11

PAGE 12

MCU microcontroller unit MEMS microelectromechanical systems PCB process controller board PDIB power-distributionand-integration-board PID proportional-integral-derivative PPS pulse per second PWM pulse width modulation RAM random access memory RISC reduced instruction set computing RISE robust integral of the sign error RMS root-mean-square SD sandisk SRAM static random access memory UAS unmanned aerial systems UAV unmanned aerial vehicle UF University of Florida USART universal asynchronous receiver/transmitter USB universal serial bus UTM universal transverse mecator WAAS/DGPS wide area augmentation system /differential global positioning system 12

PAGE 13

Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science DESIGN AND VALIDATION OF AUTO NOMOUS RAPID MAPPING SYSTEM USING SMALL UAV By William Scott Bowman August 2008 Chair: Peter Ifju Cochair: Ahmed Mohamed Major: Mechanical Engineering Advancements in inertial measurement, gl obal positioning, and digital photography have allowed the development of small autonomous unm anned aerial vehicles (UAV) to be used for low altitude airborne imaging. The University of Florida has long been interested in using UAVs for wildlife and natural re source management. Digital, high resolution images taken from the UAV can be arranged to create a map of the flight coverage area. These maps can express a multitude of various environmental constituents that can be useful to biologists and ecologists alike. Currently, aerial photography and mapping is generally done from a large fixed wing manned aircraft. Creating georeferenced maps fr om these photos is ve ry expensive and time consuming, restricting the freque ncy in which the aerial photos ar e taken. Natural resource and disaster relief managers often will sacrifice accur acy for affordable, near real-time georeferenced assessment of the surveyed area. This thesis presents a low-cost, end-to-end system for rapid single image geo-registration of airborne imagery collected from a small autonomous unmanned aeria l vehicle. Custom electronic hardware is develope d to facilitate easy user interaction with the UAV as well as 13

PAGE 14

14 INS/Camera synchronization to perform direct ge oreferencing. A method of field calibration for the system is devised to provide fast, repeatable results. Mapp ing software called PolarisView was developed to register direct ge oreferenced images into Google Earth post-flight in a rapid, near-real time manner. A test flight over pre-surveyed targets is condu cted to access the initi al accuracies of the direct georeferenced solution. By comparing the coordinates of th e targets that lie within the overlaid Google Earth image to that of their actual lo cation, a measure of residuals was established. A 34 picture subset was selected for spatial comparison; a total of 62 instances of the control points were capture d. It was determined that 4.9 % of the control points were identified within a 25 meter radius of their actual location, 36.5% within 50 meters, 27.6% within 75 meters, 21.3% within 100 meters, a nd 9.8% within 150 meters. Therefore, a 67.62 meter RMS error exists in the direct georef erenced image solution across all measurements.

PAGE 15

CHAPTER 1 INTRODUCTION In many natural resource and civilian applic ations, frequent georef erenced high-resolution imagery is useful in answering numerous questions Traditionally, satell ites and specially-suited manned aircraft provide these services. Howeve r, for many reasons these technologies fall short of fulfilling the flexible, quick-response needs of natural resource management and disaster relief personnel. Therefore, a supplemental aerial ma pping tool is necessary. Unmanned systems have proven themselves to be capable of achieving similar objectives as their manned predecessor and show much potential in filli ng the void in existing aerial ma pping technologies. A small, inexpensive, quick-to-fly Unmanne d Aerial Vehicle (UAV) could be the solution. Most small UAVs are used for surveillance and reconnaissan ce only and are not designed specifically for high resolution mapping applications. Researchers at the University of Florida are one step closer to being able to produce frequent, high re solution geo-registered still imagery from a small UAV. Motivation With the onset of personal computers and th e internet, digital ma pping technology is used on a regular basis. Maps and ge oreferenced information is accessi ble to a variety of people who need to make quick, informed decisions on a vari ety of issues; from emergency relief efforts to long term ecosystem management. Many of these ma ps are out of date and of low detail due to cost restraints within civilian funding agencies. Small Unmanned Aerial Systems (UAS) have been developed from high dollar military budgets for the last twenty years; consequently, this spending has provided the means of affordable tec hnology that will be benefi cial to an array of civilian agencies. First and foremost, UAVs can be used in certain applications to remove the element of risk from the human pilot in treacherous environments. 15

PAGE 16

Traditional low altitude natural resource asse ssment is very costly and dangerous, but necessary in many situations where ground-ba sed sampling is troublesome. Using aerial photography can reduce costs by as much as 35% for the mapping, inve ntorying, and planning involved in the management of forest and ra ngelands [1]. Natural resource managers and biologists are among many that l ook to benefit from civilian UAS technology. Biologists put their life on the line frequently to take aerial surveys that UAV t echnology could easily replace. In fact, small aircraft crashes are the leading cause of work-related death for wildlife biologists. From 1937 to 2000, 66 percent of fatal accidents were contributed to small aircraft incidentals [2]. Just recently (March 2008), a small Cessna crashed killing the pilo t and three biologists conducting a wading bird survey in the Everglades [3]. Just two months later, a similar wading bird survey was performed by a UF UAV. The University of Florida (UF) has been wo rking for six years on a simple, inexpensive, hand-launchable UAV for wildlife and natural resource monitoring. This format UAV allows for easy storage, quick accessibility, and relatively sa fe operation for civilian use. Further, recent advances in digital imaging allow UAV platforms of this size to produce comparable quality images to that of low altitude man-collected imagery. In 1999, the College of Natural Resources and Environment at the University of Florida initiated the UF Wildlife UAV program by conducti ng a feasibility study using a UAV to survey wildlife in a variety of habitats [4]. Several make-do and commercial off-the-shelf (COTS) platforms were tested and evaluated for their aer ial imaging capabilities. These aircraft used progressive scan digital video cameras whic h produced decent imagery, but not of high resolution or georeferenced. Also, both aircraft us ed an unreliable nitromethane-gas engine that 16

PAGE 17

ended up being the cause of failure and the surr ender of both platforms. A better UAV platform was essential. The Micro Air Vehicle Laborat ory, part of the College of Mechanical and Aerospace Engineering, became collaborators of the Wildlife UAV program in 2002 and produced a much improved, specially designed electric-powered co mposite aircraft called the Tadpole. This platform was outfitted with be tter avionics and imaging capabi lities [5]. However, improved sensor integration was necessary to acquire georeferenced imagery. The Geomatics program at the University of Florida became involved in the program in 2005 and began working with image post-processi ng and georeferencing. The first application of the captured imagery from the UAV was to count populations of birds by an automatic detection algorithm [6]. Later, a study was pe rformed over the National Bison Range to assess video georeferencing capabilities of the UAV. It was determined that in order to successfully georeference video frames, the platform needed: a better calibration procedure, a more accurate GPS/INS system and a better method for synchr onizing UAV attitude and image exposure [7]. The Polaris UAV Overview Since early 2007, the UF Wildlife UAV program has been focusing on many of the issues past literature expressed as ch allenges with a UAV platform of this size. The latest UF iteration, the Polaris UAV, addressed previous s hortcomings by adding the elements of aquatic recovery, easy user control and directly georef erenced imagery. A parallel effort on a levee monitoring project for the Army Corps of Engin eers inspired the development of the Polaris UAV. The 2004 version of the University of Flor ida airframe served as a base design for the Polaris since it had proven itself in to ugh operating environments [5]. The Polaris UAV was designed with a larger wing area in order to accommodate heavier camera payloads and a slower cruise speed. Th e Polaris uses simple full-function conventional 17

PAGE 18

control; throttle, ai leron, elevator and rudder. The Po laris UAV is an autonomous imaging system, meaning the aircraft is capable of tracking attitude, altitude, airspeed, and course without relying on human input for control. While unde r autonomous operation, the aircraft can perform synchronized digital imaging. Power management circuitry was devised to eliminate the need for multiple batteries. All subsystems are now interfaced using customized circuit boards with reprogrammable microcontrollers; this allows for payload synchronization and control as well as mission specific changes and further development. The airframe and electrical components were made waterresistant for UAV water recovery. A condensed version of the UAV ground station was conceived with dual touch screen monitors to be able to view all UAV contro ls and video output. A Windows Visual C++ user interface was designed to aid the user in inerti al sensor calibration, payload camera interfacing, and exterior lighting control. Further, this soft ware allows the user to control the synchronized automated picture capture function of the Polaris. Output from the UAV includes synchronized dig ital images and aircra ft state variables on two independent portable SD memory cards. The dual memory card method reduces down time between UAV flights; the user has to simply replace both SD cards and main battery to initiate another flight. The UAS was dubbed Pol aris (as in the Polaris star) because of its bright white forward pointing LED. Georeferencing and Rapid Mapping In many instances, quick, moderately-accuracy single image mapping outweighs the long turnaround time of high spatial accuracy mosaics; assessment of levee structural conditions is one example. In order to achieve comparable resolution of expensive high altitude aircraft, small UAVs must compensate by flying closer to the gr ound, therefore reducing their field-of-capture. 18

PAGE 19

19 Time and labor involved collecting ground control points necessary to improve the accuracies of small footprint photos through indi rect georeferencing contradict the usefulness of the quick-tofly abilities of the small UAV. Further, most areas in which UAVs are useful are nearly inaccessible to place ground contro l points. Therefore, the Pola ris UAV uses a newer concept of direct georeferencing to determine phot ographic area extents. Unlike many small UAS that only use wirelessly transmitted video for imaging, the Polaris uses a modified COTS digital frame camera to collect high-resolution images during autonomous operation. The commercial point-and-shoot cameras are not metric imagi ng devices and require additional calibration steps to be useful in ma pping applications. A calibration procedure is developed through a well documented Matlab camera calibration toolbox [8]. The COTS software does have the ability to ha ndle direct georeferen cing input, but often requires a great deal of human interaction to pr oduce usable results. Further, commercially available software limits flexibilities in adapting to advancements in the platform. Consequently, a Matlab-based graphical user interface called PolarisView was developed to handle the postprocessing and mapping of the Polaris output da ta in an automated, rapid fashion. The formatted, georeferenced images are importable in to a geographic information system (GIS) such as Google Earth for further spatial analysis.

PAGE 20

CHAPTER 2 LITERATURE REVIEW UAV technology has proven itself very useful in the Iraq battlefield as well as in other hostile environments. The same capabilities of unmanned systems used in warfare will contribute to a multitude of civilian applications in the coming years. Each size class of UAV has its own niche in the spectru m of mission requirements. For many reasons, the small electric UAV continues to be the favorite amongst wildlife biologists and ecologists. The advancement of imaging systems and power generation t echnology will pave the way for the future generations of small unmanned aerial mapping systems. Small Unmanned Aerial Vehicles The majority of the research and developmen t of small unmanned aeria l vehicles has been in the military context. However, their successes have spurred many new applications in other civilian interests. Configurati ons and sizes of UAVs vary greatly with application, but the focus here is on small, fixed-wing unmanned aerial vehicles. A small UAV is defined by the army as havi ng less than a 4 meter wingspan and weighing less then 55 pounds [9]. The term Unmanned Aeri al Vehicle has been recently renamed by the Department of Defense and the Federal Avia tion Administration to the Unmanned Aircraft System (UAS), but is used interchangeably thro ughout this paper [10]. The inception of small UAV technology began with the creation of the Aero Vironment Pointer in the late 1980s [11]. UAS technology remained somewhat undeveloped until 1996, when Defense Advanced Research Projects Agency (DARPA) launched it s Micro Air Vehicle (M AV) program. In 2001, DARPA began focusing on the mission capability as pect of MAVs and funded many projects that produced tiny aircraft such as the AeroVi ronment Black Widow to be used in covert military operations [12]. Academic ventures such as the International Micro Air Vehicle 20

PAGE 21

Competition sought to unveil new methods and technologies for creati ng these undetectable micro aircraft. The University of Florida has been extremely successful in this competition, winning seven of the eight years they compete d. Compliant, wind gustalleviating composite aircraft developed at UF was the contributing factor to numer ous successes [13],[14], [15]. Similar technology has been implemented on deployed small UAV platforms used by the military [16]. Research efforts eventually shifted back from MAVs to small UAVs due to the very limited payload capabilities and endurance. The primarily use for small UAS platforms has been for surveillance and reconnaissance, however pa yload capabilities of this size class UAV are attracting use in other remote se nsing applications. Many academ ic institutions are assessing a variety of research topics that will benefit many civilian prosp ects [17],[18],[19]. Small UAVs have been used for many different precision agricultural applications such as vegetation monitoring and crop yield estimation [20], [21], [22], [23] UAS have even been used for more exotic exploratory endeavors such as volcanic gas sampling, forest fire monitoring and hurricane research, where aircrew of manned missions were at the greatest risk [24],[19], [25]. Aerial Imaging and Georeferencing Many different kinds of imaging payloads exist for small aerial vehicles. Common imaging sensors used in modern small UAVs in clude visible spectrum, infra-red and thermal infrared imagers. All have diffe rent optics and measurable wavele ngths, but still adhere to the same photogrammetric principles needed for georeferencing. Two forms of georeferencing exist for compu ting desired spatial relativities. Indirect georeferencing uses ground control points to back-calculate the external orientation parameters of the imaging sensor [21]. Direct georeferen cing estimates ground coordi nates based on attitude and positional measurements made on the aircraft [26]. 21

PAGE 22

Several methods of indirect georeferencing exist and are used widely among UAS. One such process called image-to-image georef erencing uses common tie points between two images to register one image to the other [ 27],[7]. Image-to-Map georeferencing uses identifiable ground control points in each imag e to perform a transformation from pixel row/column information to aircraft orientation needed for locating other ground locations in the image [28]. Another form of indirect georef erencing is calle d aerial triangulation. This procedure uses identical object points in two separate photogra phs to determine height and position estimates [29]. Directly georeferenced images require synchr onizing the timings of sampled position and orientation estimates of the aircraft with the cameras exposure [18]. Rigid body transformations of the imaging plane element projections a nd intrinsic camera para meters allow ground coordinates to be determined [26] Attitude and position estimates used for direct georeferencing in small UAVs are generally provided by their navi gation instrumentation. Integrated low-cost global positioning system (GPS) and microelect romecanical systems (MEMS) based strap-down inertial measurement units (IMU) are popular guidance systems amongst small UAVs. This technology is used for the benefit of reduced si ze and weight; however forfeits position and attitude accuracies. Boresight, the rotation of the imaging sensor versus the IMU and lever arm offsets, distance between imaging sensor and IMU, must be taken into account in the georeferencing solution [30]. Small UAVs are lightweight, low altitude aircra ft and are therefore susceptible to wind turbulence and high dynamic situations rendering attitude estimation and synchronization nontrivial. Atmospheric conditions are more suitable for vertical imaging from high altitude UAVs, however, cloud cover at these altitudes can cause obstructed landscape. Few small UAS in 22

PAGE 23

literature use direct georeferen cing, however, much faster mapping results can be achieved when low-accuracy is acceptable [29]. High accuracy rapid direct geor eferencing systems due exist on sizable manned aircraft but have not been refi ned for smaller, less-accu rate UAV systems [31]. Both forms of georeferencing re quire the determination of in terior and exterior camera parameters. Interior constituents consist of camera parameters such as effective focal length, scale factor, principle point offset and lens distortion [32]. Exteri or parameters are a function of sequential rigid body rotations describing the rela tionship between the image plane and the world coordinate frame. Many methods for determining interior and exterior cam era parameters exist, but all have a similar underlying process; satisfy the collinearity condition based on known spatial locations. The collinearity condition guar antees that at the e xposure station, an object point in the mapping frame and its photo image in the image plane all lie along the same line in three-dimensional space [7]. The selected pr ocess for determining the camera sensor model used for the Polaris platform is very comprehens ive and includes both radial and tangential lens distortion components that can be used to correct image distortion and projective geometry extents [33]. Digital Mapping Projective geometry is used to predict image c overage extents and is a function of tilt about all axes of the aerial platform [34]. However, acquiring the ground coordinates of the captured area is only a start in the mappi ng process. A geometric distor tion exists in any non-vertical photograph and must be removed before the image can be registered into a reference datum [26]. Geo-registration is the process of flattening tilted images until their normal vector is parallel to the mapping systems normal vector in a georeferenced location [27]. Differential orthorectification accounts for di stortions in the photo caused by he ight differences of elements that span a photograph [34]. This case is not considered for Polaris image mapping, but is a 23

PAGE 24

24 focus for future research. Aerial imaging platfo rms using film cameras require mechanical and optical equipment to remove tilt distortions in the photographs, which is a very time consuming manual process [34]. With the inception of di gital imaging, these processes are no longer necessary. Digital rectification re quires the use of dig ital image processing to resample the raster data to create an undistorted di gital image [35]. The resampling process alters the row/column makeup of the digital image based on projective geometry constrai nts to allow the output image elements to cover equal units of area; as if the image was taken vert ical to the ground. The increase in computing performance over the la st ten years allows for high pixel-count (high resolution) digital images to be rectified quick ly. The assembled map can then be used for various ecological photo interp retations such as grid sampling, where information about a viewed area can be extrapolated to pr edict tendencies in larger areas [27].

PAGE 25

CHAPTER 3 POLARIS UAV AIRFAME The University of Floridas Micro Air Ve hicle (MAV) Laboratory has been developing small and micro unmanned air vehicles for almost twelve years now. The laboratorys expertise in composite materials and manufacturing has al lowed for the development of some of the smallest and lightweight flying planes in the world. The design and construction methodologies of the MAV laboratory has always been iterative and geared toward rapid, inexpensive, mission capable composite aircraft that ar e easy to fabricate and assemble. Past UF UAVs were driving factors in the Polaris design. The followi ng gives a brief overview of the design and construction of the Polaris airframe as well as the electronic equipment associated with the current technology. Construction The Polaris UAV demanded more unique marine operating environments then its predecessors, so several different construction materials and techniques were employed. Polaris design was easy to construct for someone with intermediate level of composite material experience. The UAVs conventiona l layout and modular design allo ws for fast field assembly and easy replacement of damaged parts. The fuselage design of the Polaris UAV remain ed identical to the previous UF design (Figure 3-1) [5]. The Polari s UAV was designed using lightwei ght materials and methods to keep overall airframe weight to a minimum. The fuselage was constructed out of 189.9 grams/sq meter woven Kevlar fabric a nd a high strength epoxy-based re sin rendering an exemplary strength to weight ratio and ve ry abrasion resistant. The motor mount portion of the fuselage was reinforced with a 193.3 grams/ sq meter carbon fiber cloth fo r rigidity. The fuselage was dunk-tested to ensure it was free of water leaks; any voids were patched with a fast drying two 25

PAGE 26

part epoxy. A fiberglass hatch wa s made from a rubber-plugged female mold that was machined from high density foam tooling-board. The wing and tail sections of the Polaris UAV were made of a fiberglass wrapped lowdensity Styrofoam and epoxy-doped balsa stock. Wi ng and tail section designs were sent away to Flyingfoam.com to be hotwire cut out of large blocks of 16.0 kilogram per cubic meter EPS foam. This foam was selected for its spar se closed-form porous surface allowing adequate infusion of epoxy from the wrapped fiberglass skin; this keeps the fiberglass cloth from delaminating from the foam. Table 3-1 shows all wing and stabilizer design parameters. Figure 3-2 shows the planform of the wing and stabilizer sections The wing sections had three holes and span-wise rectangular grooves made during the hotwire cutting process. Two of the holes were cut mid-span into the wing and were reinforced with cardboard tubing to accommodate the carbon fiber wing joining spars. The largest hole for the load bearing spar was 21 millimeters in diameter while the torsion tube hole was 14 millimeters. The third hole was cut full span length at center chord to accommodate the electrical wiring throughout the wi ngs. Span-wise grooves facilitated square 4.8 millimeter balsa stock to aid in bending rigidity. Epoxy-doped spruce hardwood was placed in critical areas in the wing and stabilizer sections for fast ening and mounting. Control surfaces were cut from the EPS blanks and hinged with a rectangular piece aramid fabric. All wing and stabilizer sec tions were wrapped with 115.3 grams/sq meter s-glass (structural fiberglass) fabric then wet with a sl ow cure, high strength epoxy resign and tensioned with low-density acetate release film for a nice surface finish (Figure 3-3). Wet sections were then placed in their respective female portion (shuck) of the orig inal EPS block to keep shape while under external weighting (Figure 3-4). After a full twenty-four hour cure, hardened 26

PAGE 27

sections were removed from their shucks and t ouched-up with a lightweight, foam-safe filler to fill any pinhole water leaks. Assembly All fuselage, hatch, wing and stabilizer secti ons were sanded and painted their respective coloring; blue and orange of course. An epoxy-doped piece of 1.5 mm spruce hardwood shaped as the cross section of the fuselage was glued in to place to serve as a primary bulkhead for initial strengthening. A 19 millimeter diameter, 1 meter long carbon fiber tube was fitted through the back of the fuselage and mated with an accommod ating hole in the center of the bulkhead inside the fuselage. The carbon tube was glued in to place at all contact points using Hysol adhesive. Holes were cut out of the fuselage for the fo rward pointing white LED (light emitting diode), female camera lens insert, speed control, video transmitter, forward looking camera, pitot tube and on/off switches. Stainless st eel 3 mm screw/nut pairs were installed on the four corners of the hatch opening to server as o-ring posts to hold down the hatch during flight. A receiving saddle for flush wing/fuse in teraction was made out of fibe rglass. Two 9.5 mm carbon fiber dowels were installed approximately 28 millimeter s below the saddle to serve as rubber band wing fastener posts. A 6.4 mm car bon tube along with plastic tubi ng was installed in the nose of the fuse to serve as a Pitot tube. Plastic t ubing was connected to a yjunction plastic housing with water catch to keep water away from the dynamic pressure sensor of the autopilot in the instance of a water landing. A 304 mm section of the carbon fiber tube was cut off the back of the tail boom to be used for stabilizer mounting. The moto r was installed onto the fuselage with stainless steel hardware and the motor wiring was sealed with a rubber grommet and ran into the fuselage. Marine grade weather stripping was installed around the wing saddle and main hatch lip to form a water resistant seal when hatch and wings are installe d. A carbon fiber sheet wa s laid up and cut into 27

PAGE 28

an ovular shape to fit snuggly in the upper third of the fuselage; this formed a hatch that was siliconed into place to keep water out, but could be removed if necessary (Figure 3-5). The cable for the wings and a programming header was instal led here in this saddle hatch. Two on/off switches were rubber booted and in stalled into the back of the fuselage above the tail boom for easy access. Wing and stabilizer control surface areas were scored and cut appropriately to allow adequate deflections. Titanium control clevises with plastic stop horns were installed on all control surfaces. Servos were mounted in all EPS wing and stabi lizer cores using a plastic servo mounting tray that facilitated eas y servo replacement. Servo hatc hes were devised out of flat pieces of fiberglass that had rectangular holes made in them so that a rubber servo arm boot could be installed to isolate water from the servo chamber. Four 12.7 mm neodymium magnets were installed into the root of each wing to serve as wing joiners. Red and Green LEDs with complete wiring were installed at the tips of the wings and a white LED was installed at the top of the ve rtical stabilizer. All LEDs were covered with a plastic globe for waterproofing. Guides were cut in the trailing edge of the wing near the root to accommodate the rubber bands used for wing attachment to the fuselage. Holes were drilled into hardwood sections of the horizontal stabilizer so it could mate with aluminum tail boom mounting hardware (Figure 3-6). Two vertically mounted 5 mm carbon tubes were ran through the horizontal stabilizer and were used to s upport the vertical stabilizer to the separated tail boom section. An access hatch wa s cut out of the vertical stabilizer for radio modem installation and connectivity. A carbon tube fitting the inside diameter of th e tail boom was glued into place to serve as a coupling between tail boom and tail section. A 6.4 millimeter stainless steel screw/nut set was 28

PAGE 29

used to fasten the tail to the tail boom. A cap was made for the open end of the tail section for waterproofing. Figure 3-7 shows a completed Polaris UAV airframe. Initial flight testing was conducted with just the airframe and COTS radio-controlled electronic hardware. The plane was outfitted with early generations of the customized electronic hardware to collect and wirelessly transmit power draw and temperature readings of critical flight components to ensure reliability for real missions. The initial flight platform proved it could fly for 50 minutes on a single 10000 mAH, 18.4V Lithium-Polymer battery. This was the battery chosen to be used in actual missions. However, a smaller capacity, lighter battery could be used to offset the weight of heavier future pa yloads. Mission flight time s were restricted to 40 minutes for safety. It was determined that an airspeed of 14 meters/second was as safe flying speed for the aircraft and provided a slow e nough ground speed for sharp images. A critical airspeed of 10 meters/second would cause the aircraft to stall. The finalized aircraft conducted missions in the actual habitats it was designed for. Mission take-off scenarios including land, motor boat and airboat were all very successful (Figure 3-8). Landings were accomplished on land, in open water, and within marshy, cattail lined wetlands. Day and low-light missions were successfully tested as well (Figure 3-9). Protocol for mapping missions entailed manually fl ying the plane from hand-launch up to a safe altitude (approx. 100 meters) then engaging autopilot control for the majority of the flight. During this time, flight coverage area, altitude or payload set tings could be altered using the ground station interface. Also, the UAV could be toggled in and out of synchronized image capture mode. After the survey was complete d, the Polaris was landed using manual control. 29

PAGE 30

Table 3-1. Wing and stabilizer design parameters Section Airfoil Span Root Chord Tip Chord Sweep Washout Wing NACA2313 2.4384 m 0.3048 m 0.2032 m 1.19 deg 2.5 deg Horizontal Stab NACA0012 0.6913 m 0.2095 m 0.1168 m 3.83 deg 0 deg Vertical Stab NACA0012 0.36957 m 0.2286 m 0.1295 m 3.83 deg 0 deg Figure 3-1. Unpainted Kevlar Fuselage used for the Polaris UAV Figure 3-2. Polaris wing, horizontal and vertical stabilizer plan forms 30

PAGE 31

Figure 3-3. Tensioning acetate film over fibe rglass skinning creating a smooth external surface. Figure 3-4. Process for compressing fiberg lass curing epoxy to EPS foam wing cores 31

PAGE 32

Figure 3-5. Saddle hatch for programming access of PDIB processor and wing connectivity Figure 3-6. Horizontal stabiliz er tail boom mounting hardware A B Figure 3-7. Completed Polaris UAV airframe A) Birds eye view B) side view 32

PAGE 33

33 Figure 3-8. Airboat launch and recovery set up. The mission was conducted in an Everglades marshy environment. Figure 3-9. Polaris day and lo w-light flying configuration

PAGE 34

CHAPTER 4 POLARIS UAV ELECTRONICS The end goal of the development of the Polaris UAV was to keep everything as inexpensive, reliable, simple, and hand-launch able as possible in order to conduct aerial surveys. Since the Polaris is of the same size as many model aircraft, quality COTS hobby components were used in order to reduce time-to -flight as well as cost. Some of these components were slightly modified in order to m eet our goals, such as waterproofing. However, when it came to power management, synchrono us imaging, data collection and systems integration, there were no solutions available in our size, weight, and cost budget. Therefore, custom electronics were developed for special use on the Polaris UAV. All custom printed circuit boards are two layer boa rds and were designed using Altium Design Explorer. All printed circuit boards were exported to Advanced Circuits for manufact uring and populated at laboratory facilities at UF. Every form of el ectronic circuit board wa s encapsulated with MG Chemicals brand encapsulating compound to ensure water-resistant, vibration proof operation. Certain components were covered with thermally conductive encapsulating compound depending on heat dissipation requirements. Motor Selection and Controller Previous versions of the University of Florida UAV used a nitromethane internal combustion engine that proved to be very messy, noisy, and unreliable. The prequel to the Polaris had an electric motor th at performed very well [5]. An E-Flite 46 860 KV, brushless outrunner motor was selected for the Polaris b ecause of its increased torque and improved cooling. Also, it is inherently waterproof in operation; a great adva ntage over brushed motor technologies. The basic concept of brushless mo tor technology involves using variable induction rather than electro-mechanical brushes for torque generation. There is no electrical contact to 34

PAGE 35

short in case of water recovery. (The motor bearings and iron cores were sprayed thoroughly with silicon-based lubrication after each water landing.) A Jeti 70 Amp brushless speed control was select ed to control the UAVs electric motor. This speed control surpasses the maximum ampera ge draw for the motor considerably as to ensure reliability. The entire speed control was covered in a thermally conductive epoxy and a 50.80 mm by 13.46 mm by 4.83 mm finned heatsink was added to ensure reliable performance. The motor/speed control combination produces approximately 1025 watts of power with a 10x8 three-bladed propeller, which produces ade quate hand-launchable takeoff thrust. Actuators The Polaris UAV uses servomechanisms (servos ) for actuation of control surfaces. Wing and stabilizer sections were designed for the serv o to be installed within them; therefore a lowprofile, thin servo was required. The JR model DS168 digital thin-wing metal-gear servo was selected for the Polaris UAV. The DS168 is 6-vo lt tolerant and has subs tantial torque for its size. The servo exhibits .374 Newton-meter of to rque at 6 volts and travels 60 degrees in 0.14 seconds. Autopilot The UF MAV Laboratory has been working with the Procerus Technologies autopilots through several of their iterations. The latest ve rsion of the Kestrel auto pilot has proven itself very reliable and was therefore selected for use on the Polaris UAV. Offering a rich feature set, the Kestrel v2.23 makes integration a nd setup a seamless process [36]. The Kestrel v2.23 weighs only 16.65 grams and occupies less than 21.11 cubic centimeters (Figure 4-1). The autopilot uses a temperature compensated MEMs-based three-axis gyroscope/accelerometer as the iner tial measurement unit (IMU). The autopilot has a three-axis magnetometer and GPS for navigation. It has been reported in literature that this specific 35

PAGE 36

INS/GPS navigation suite is capable of accuracies shown in Table 4-1 in near level conditions with post-processing [30]. The Kestrel v2.23 has 3 servo and 1 speed contro l output. A barometric pressure port is used for altitude estimation while a differential pressure port is used for airspeed estimation. The Kestrel v2.23 also has an open configurable serial port for external payload interfacing. Customized circuitry developed for the Pola ris UAV is interfaced through this port. A Maxstream 1 Watt 900 MHz radio modem is used to communicate with the ground station and is mounted in the vertical stabilizer to a void electromagnetic interf erence (EMI) with other sensitive electronics (Figure 42). The autopilot was encased in a carbon fiber box for protection and waterproofing (Figure 4-3) External connectors provid e signal and power routing. The autopilot uses a combination of feedback and feed-foward cont rol to stabilize and navigate the aircraft. Proportio nal-Integral-Derivativ e (PID) feedback cont rollers are used to regulate aircraft states and a feed-f orward effort is applied in certain scenarios such as turns. All control loops have user define d saturation limiting to ensure ai rcraft stability and undesired aerodynamic loading. A real-time PID data logger in the user-interface was used for tuning the PID controllers while flying the UAV for the first time. Figures 4-4 through 4-6 show all control loop structures used in the Kest rel v2.23. During strai ght flight paths, th e autopilot does an exceptional job holding the aircraft to its calibrated le vel condition. Only dur ing situations of external excitation, such as turbulent wind, do the controllers have a difficult time commanding the UAV to remain level. Global Positioning System Receiver The Global Positioning System is the critical element of the INS/GPS integration that allows small UAVs to navigate with reasonable ac curacy. For the size clas s of the Polaris, many COTS receivers are available. However, only cert ain units are compatible with the autopilot. 36

PAGE 37

The GPS receiver used on the Polaris is a Furuno GH81 model (Figure 4-7). This model is a low-power, 3.3 volt, 16 channel GPS with 15 mete r horizontal accuracy (2drms) and 22 meter vertical accuracy (2drms). The GPS receiver is actually mounte d on a printed circuit board and ground plane provided from Procerus. The receiver is m ounted on the right wing 40 centimeters from center. The GPS receiver is covered with a non-conductive plastic housing for splash resistance. Subsystem Control Hardware Synchronization amongst the payload imaging se nsor and the autopilot are necessary to accomplish direct georeferencing. Further, the abili ties to adjust functions of the payload while in flight were necessary. To integrate the payload sensor and other Polaris subsystems, customized electronics were required. Since the decision was made to retain th e fuselage from the Tadpole design, special considerations for physical circu itry layout had to be considered in order to remain within the physical constraints of the fuselage. The subsystem interface hardware, named the powerdistribution-and-integration-board (PDIB), was designe d to have a multipart role in the Polaris. The PDIB was fabricated on rigid, 1.6 millimeter thick FR-4 substrate and served both as a functional printed circuit board and as a secondar y bulkhead for further fuselage reinforcement. A hole in the center of the PDIB allowed it to ma te with the internal portion of tail boom and pass electrical cabling to the tail. The outer most boundary of the printed circuit board was designed using AutoCAD, a computer-aided design (CAD) tool, to match that of the cross-section (bulkhead) of the Kevlar fuselage. The CAD outline was imported as a keep-out-layer into the circuit design and boardlevel layout suite called Design Explorer, created by Altium. The PDIB operates as an embedded system with an 8 Mhz Atmel Atmega 128 8-bit RISC micr ocontroller as the central 37

PAGE 38

controlling unit (Figure 4-8). Si nce the PDIB board was secure d into place, the In-SystemProgram feature of the Atmega allowed for the microcontroller to be re-programmed without having to remove any other peripherals. A pr ogramming port was installed via a RJ-11 jack located near the top of the fuselage in the saddle hatch for easy accessibility. The microcontroller unit (MCU) has many features that can be seen in [37]. The Atmega 128 MCU was particularly selected because of the dua l programmable universal synchronous/asynchronous receive/transmit serial ports. These serial lines were used to interface the payload, the autopilot and the SD card writer/reader, which made camer a/INS/GPS synchronization possible. The USART1 receive port of the MCU accepte d signals from the autopilots SERIAL-A passthrough port via a 38400 baud, 8 bit word length, 2 stop bit serial stru cture. The USART1 transmit port was used to generate 57600 baud American Standard Code for Information Interchange (ASCII) messages to the SD card writer. USART0 receive port listens to 115200 baud serial messages sent from the autopilot to the wireless modem containi ng aircraft state data used for camera synchronization. Lastly, the USAR T0 transmit line (57600 baud) is used to talk to the Atmel Atmega8 MCU on the custom camera inte rface board or with any other payload that uses serial communication. The MCU input/output (I/O) was also used to control the duty cycl e (on-time versus offtime) of the Phillips Luxeon 1 Watt low-light operation LEDs via power transistors. Digital I/O toggles a video multiplexer that controls signal routing from the payload camera and forward-looking CCD camera to the wireless video transmitter. The MCUs 10 bit Analog-toDigital (ADC) ports were used during UAV testing for data collection of various external analog sensors. During normal autonomous imaging ope ration, the ADC samples a variety of voltages 38

PAGE 39

in the design to check for power management failures. Ten additional digital I/Os were pulled out for future expandability. Data Storage Usually more then 400 high resolution digital images are taken every Polaris flight. These digital images are 12.1 mega pixels and can take up as much as 6.2 megabytes in file size, which equates to approximately 3 gigabytes of im aging data. This large amount of data could take as much as twenty minutes to transfer to the ground station computer before another flight could commence. Therefore, inexpensive porta ble memory was chosen over a USB 2.0 transfer method. Wiring was run from the payl oad box to an externally mounted SD memory card holder so that the cameras built -in memory management features could be used. After each flight the memory card was removed and replaced with a formatted card of the same size. UAV state data storage was accomplished in a similar fashion. A lightweight memory card management module with a DosonChip was selected to write serial ASCII data from the Atmega 128 (Figure 4-9). The DosonChip allows for easy memory card management having file structures similar to MS-DOS. The card has poor throughput, but was adequate to handle the 200 bytes/sec needed. Every time a picture is taken, a single file called, PIC_DATA.txt, is appended with space delimited UAV state data and terminated with a carri age return. A typical text file size for a 400 picture mission is roughl y 70 kilobytes, so only a small SD card (250 MB) was used. Other articles of data, such as camera calibration and positional parameters, share space on this card for digital mapping purposes. Cameras The Polaris UAV has two cameras on-board for different purposes. The first camera is a CCD chip camera used as a situational awarene ss camera giving the oper ator a first-hand look through the front of the UAV. The second cam era was the payload camera whose type was 39

PAGE 40

dependent on mission application. All testing with the UAS has been with a digital camera payload for direct georeferenci ng testing, however, other sensors, such as a thermal-infrared imager is awaiting installation. The CCD chip camera circuitry was sealed wi th encapsulating epoxy and installed in the nose of the Polaris (Figure 4-10). The 5 vo lt, KX-141, 13 gram, 640 x 480 resolution, color CCD video imager was selected as the forwar d looking camera (Figure 4-10). The NTSC output of this camera was multiplexed with the payl oads camera NTSC output for toggling of live video via a 1 Watt wireless 2.4 GHz transmitter. Live forward looking video also provided visual aid in control loop tuning and manual la nding exercises. The transmitter and custom video control circuitry genera ted large amounts of heat, so they were encapsulated in thermally conductive epoxy and mounted on a common external heatsink for heat dissipation (Figure 411). The digital imaging payload used in th e UAS was a COTS Canon A650 12.1 mega-pixel point and shoot camera which provided 25 mm pixel ground coverage at 100 meter altitude. The camera was stripped of all non-imaging elemen ts such as the flash, battery holder and any other unnecessary casing, shaving off about 35% of its total weight (Figur e 4-12). A lightweight, carbon fiber housing with 52 millimeter lens cove r was made to protect and waterproof the cameras internal components. The housing was lined with low-density foam to isolate the sensor from vibrations. Originally, the use of an IEEE1394 protocol for controllability of the A650 was tested, however, it was not possible to control the manual focus functionality of the camera. A manual focus setting at infinity reduces static shutte r lag compared to the dynamic behavior of the 40

PAGE 41

cameras built in auto-focus function. A PCB was designed with an Atmel Atmega 8 MCU to control the Canon A650 digital camer a hardware (Figure 4-13) [38]. The custom camera controller board received serial comman ds from the Atmega 128 on the PDIB to control all aspects of the camera, including going in between still picture and video mode. Digital I/O signals were run from the Atmega8 MCU to all input buttons of the Canon A650 hardware. Insulating Kapton tape was used to secure the circuit board and wiring to prevent shorting. A digital oscilloscope was used to capture the waveforms needed to trick the camera into thinking a button was pressed. The waveforms were regenerated on the Atmega8 at desired times, acting like a very fast digital finger. A source-able line driver was used to trigger the camera to turn on and off. Powe r was provided to the camera and controller board from the power module described below. Polaris Power Management The previous version of the University of Florida UAV required th ree separate battery packs in order to operate all subsystems; this became troublesome between flights when the wings had to be removed in order to change th ese batteries. Also, no voltage level indication from two of the three ba ttery supplies was available. Therefore, a power regulation design was implemented that allowed for only a single battery to be used. There are many issues that arise during efficient DC/DC power management, especially in large voltage differential situations as the case on the Polaris UAV. Multiple iterations were made to the power design and tested for airworthiness. A DC/DC step-down solution was de veloped that is flexible for future platforms and proved to be very reliable. The primary concern for the power management design involved reduc ing as much as 21.2 volts, from a fully-charged battery, down to no greater then 6 V for UAV components. Heat 41

PAGE 42

dissipation is usually a concern in this scenario. Sensitive digital circuitry was used in the aircraft, so voltage levels needed to be well conditioned. The power management design was a hybrid of switching and linear vo ltage regulators. The design is cascaded and separated into two ca tegories of electronics; noise-sensitive and nonsensitive. The front-end voltage regulation is done with a high-power sw itching regulator and is used to power non-sensitive devices such as LEDs, servos and camera payload (which has its own set of regulators). The rema ining voltage regulation is done w ith linear regulators. All the current to power the system has to be sourced from the switching regu lator which has a much better heat dissipation property to that of linear architecture. The video transmitter is the only device that uses a straight linear regulator from the input batte ry; however, it only has to step down to 12 volts instead of 6 volts. The switching voltage regulator that was chosen to do the front-end power regulation was a National Semiconductor LM2677 Series adjustab le voltage regulator in the LPP (Leadless Plastic Package) package. This regulator can source up to five amps and can accept a wide input voltage range (between 8 to 40 volts). The output voltage is adjustable from 1.2 volts up to the input voltage. The internal switching frequency is betw een 225kHz 280kHz depending on output voltage. A two resistor bi as network exists to alter the adjustable output voltage. A 5.8 volt output from the regulator was selected in or der to remain within operating limits of other electrical components. A special PCB layout was devised for this regulator so that the design could be inserted into other circuitry using 2.54 millimeter pitch male headers. The small, output-adjustable, easy mounti ng voltage regulation board was called the power module. The power module operates between 85-88% efficiency depending on instantaneous current output (F igure 4-14). A ground plane on the bottom of the board and 42

PAGE 43

aluminum rectangular shieldi ng was used to help reduce EM I of neighboring circuits. Approximately 9 Watts is available to the payl oad to keep the power module within a safe operating range. A 5 volt and 3.3 volt linear regulator was used after the power module output for more sensitive electronics where stability and voltage ripple are more critical. Both microcontrollers and the GPS receiver were pow ered with these supplies. Ground Control Station The ground control station (GCS) serves as the critical link between the UAV and ground operators. Many modifications were made to th e GCS over past years renditions for ease of use and reliability. The housing of the GCS is a watertight, wheeled Pelican case with external peripheral interfaces (Figur e 4-15). A Panasonic Toughbook is the central piece of equipment in the GCS. This laptop computer runs the Virtual Cockpit, PolarisLink and PolarisView software packages. Both monitors provide a way to distribute the multiple software GUIs spaciously so everything is accessible quickly. The GCS has a Commbox provided by Procerus Technologies with the matching 900 MHz radio modem for wireless communication and an external jack for connecting the manual pilot control box for take-offs and landings. Th e GCS uses half-wave monopole antennas for both video and telemetry reception and transmissi on. However, weatherproof N-type coaxial connectors are provided externally so that larger, more directiona l arrays can be connected for increased communication range. The Commbox interfaces the laptop computer via a standard serial cable. The GCS has a 2.4 GHz video signal receiver for displaying live video feeds from the Polaris. Video overlay of sy stem critical information is view able over this feed. The laptop has an analog video frame-grabber for recording live video streams. The GCS uses an AC/DC 140W converter to power electronic s in instances where AC power is accessible. This is useful 43

PAGE 44

for flight planning indoors. All cr itical flight electronics in th e GCS have battery-back built in. A weatherproof USB port and DC cigarette lighter port are provided for east accessibility. 44

PAGE 45

Table 4-1. Root-mean-square errors in the Kestrel v2.2 INS/GPS in near level conditions (<~4 degrees) Roll (Deg) Pitch (Deg) Heading (Deg) Northing (m) Easting (m) Altitude (m) 0.44 0.31 1.1 0.90 1.0 2.1 Figure 4-1. Procerus Autopilot Kestrel v2.2 Figure 4-2. One Watt Aerocom modem is mounted in tail to reduce EMI 45

PAGE 46

Figure 4-3. Carbon fiber box for autopilot with waterproof interfacing electrical connectors Figure 4-4. Elevator control loop Figure 4-5. Throttle control loop 46

PAGE 47

Figure 4-6. Aileron/Rudder control loop Figure 4-7. Furuno GH81 GPS receiver on Polaris UAV Figure 4-8. Custom designed Power Distribut ion and Integration Board. Figure shows placement of printed circuit board within the fuselage. Note: The Power Module is not installed in this figure. 47

PAGE 48

Figure 4-9. DosonChip SD Memory Card hardware used to store synchronized UAV orientation and location data Figure 4-10. KX-141 CCD camera and installa tion location Figure 4-11. One watt video transmitter (no ante nna) and video control circuitry. Thermally conductive epoxy was later used on the components for waterproofing. An externally mounted heatsink was used for heat dissipation. 48

PAGE 49

Figure 4-12. Canon A650 Digital Camera payload. Figure shows progression from off-theshelf to UAV ready. Figure 4-13. Unpopulated camera control board Figure 4-14. Custom designed power module. Th is picture shows the module populated with all external components. 49

PAGE 50

50 Figure 4-15. Front and back of Ground Control Station (GCS)

PAGE 51

CHAPTER 5 POLARIS UAV SOFTWARE Previous UF UAV platforms performed well fo r imaging and flight data collection, but provided no means for sensor interfacing, synchr onization or georeferencing. The Polaris UAV was designed with programmable microcontrollers that would allow for system upgrading and specific mission changes. Further, a Graphi cal User Interface (GU I) was developed for Windows that gives the user complete contro l of UAS peripherals in real-time. Firmware Firmware is software compiled to run at th e processor binary level from some higher level language. Both Atmel mi crocontrollers were programmed w ith firmware generated from the CodeVision AVR C Compiler. Each microcontroller is responsible for separate tasks, but share information amongst each other via a serial data link. A custom serial communications protocol was developed to pass data from the ground station software to the dual UAS MCUs. The primary MCU, the Atmega 128, managed a wide variety of tasks to carry out autonomous mapping functionality. This MCU wa s programmed to: issue commands to the payload, extract high rate UAV state data, generate synchronized camera shutter capture, control onboard video selection, generate pulse width modulation (PWM) signals for LEDs and write out data to the portable SD memory card (Figure 5-1). The autopilot was configured to use the SE RIAL A port as a passthrough port, enabling wireless serial messages generated from PolarisL ink to pass through this bus to the PDIB MCU. These forwarded serial messages are ha ndled by the Atmega 128 using interrupt driven receive buffers. Serial messages sent by Po larisLink are parsed in accordance to the communication protocol and corres ponding subroutines are executed. 51

PAGE 52

The firmware on the Atmega 8 handles the contro l of all digital I/O of the processor based on serial message received from the Atmega 128. Signal lines from Atmega 8 were connected to each user input button of the camera for full accessibi lity of the cameras features. The firmware also runs a startup sequence every time the came ra is turned on. This sequence sets up the cameras manual mode of operation and infinite fo cus, thereby reducing shutter-lag. After the startup sequence finishes, serial data from the PDIB is available fo r parsing to control any feature of the camera. Graphical User Interfaces Several software packages are needed in orde r to have full functiona lity of the UAV. The Virtual Cockpit is a GUI that Procerus Technologies provides with thei r product so the user has access to all functions of the autopilot. In past UAV missions it became evident that payload settings needed to be adjusted while the plane was flying. With the addition of all the on-board electronics, an easy to use GUI was developed to control all the subs ystem features of the aircraft to make this platform a truly autonomous mapping system. The Virtual Cockpit provides the UAV operators with information about various aircraft state data, critical measurements and internal autopilot variab les (Figure 5-2). The Virtual Cockpit also has a route planni ng and real-time UAV trajectory inte rface. This interface allows the user to setup waypoint entered routes for th e UAV to follow as well as displays the current position of the UAV once airborne. The UAV can be instructed to change course, altitude or airspeed at anytime during the flight. A specialized GUI was developed to enable the user to quickl y and easily control functionality of all the secondary syst ems of the Polaris UAS. A Visual C++ user interface, called PolarisLink was designed to handle this task (Figure 53). The Virtual Cockpit software for the autopilot can communicate bi-direc tionally over a local Internet Protocol (IP). 52

PAGE 53

Through this network IP, data is shared betw een Virtual Cockpit and PolarisLink. Virtual Cockpit creates special serial packets based on the custom Polaris communication protocol and appends this data to normally transmitted data over the wireless modem in the Commbox. This data is guaranteed received using an ACK/NACK system by the autopilot and then transmitted out of SERIAL A in a process described earlier. The PolarisLink GUI is very intuitive and easy to use. By selecting the Digital Camera Setup button, an additional window opens up that mimics the back the Canon A650 digital camera (Figure 5-4). By pressing these buttons, commands are transferred to the camera controller board to mimic these buttons being pressed on the ground. The wireless video disp lays all the appropria te screens, so the user can change any feature desirable while having visual feedback that it was accepted. After the camera is setup properly, the GPS Sync Shutter can be pressed to initiate the automated synchronized image capture process. This process can be turned on and off as desired. Camera Synchronization Camera synchronization refers the time at which remote sensing data is captured relative to the trajectory of the UAV [30]. Most modern GPS receivers have dedicated synchronizing signal called a PPS or pulse-per-s econd signal, which is based on GPS time and is known to be sufficiently accurate [30]. However, the GPS rece iver that was used for this paper and for the last several missions does not generate this pulse. Even if this signal was generated, synchronizing INS state data that is available from autopilot with this pulse is not possible. Therefore a different approach to ca mera synchronization was implemented. The autopilot normally wirelessly transmits state data and other autopilot variables approximately every 450 milliseconds, which is sufficient for updating the Virtual Cockpit. Every second a navigation packet (header 248) is sent containing an updated position solution 53

PAGE 54

from the GPS. Since theses packets are genera ted at different times, they cannot be used for synchronization. A special data output mode ca lled Mixed Telemetry Mode is capable of outputting data at a 6 Hz rate and contains both attitude and interpolated GPS positions. When the button on the PolarisLink software entitled GPS Sync Shutter is depressed, this special mode is enabled. The firmware in the Atmega128 MCU keeps track of how many times the packet header 29 is sent to the mode m for transmitting. As soon as the firmware sees the 14th count of the packet header 29 (~2.33 se conds), it immediately issues a camera shutter command. The time it takes for the camera to re ceive the take picture command is on the order of 15 microseconds and is therefore negligible. The remaining serial data is read into an SRAM buffer in the Atmega 128 until an end of line character is read. As soon as that buffer is full, the data is then sent to the SD card. It was experimentally determined that the camera takes approximately 1.80 seconds rootmean-square (RMS) to a store at full resolution pi cture onto the SD card. Therefore, triggering a picture capture any faster than this would cause the system to get out of sync as there is no feedback from the camera to the memory modul e. To be conservative, the system was programmed to command a picture every 2.33 sec onds. The slow flying speed of the UAV still provided plenty of overlap amongst consecutive pi ctures for constructing picture mosaics in the future. Timing delays in the exposure of the actual photo needed to be considered in the final direct georeferencing solution. Through expe rimental methods and online sources it was determined that the Canon A650 has an approximate exposure delay of 87 milliseconds. Errors can propagate in both positi on and orientation of the camer a over this time period. An appropriate method for accounting for this offs et is to integrate the accelerometers and 54

PAGE 55

gyroscopes over the delay time period and add in this bias to the original position and orientation findings at the time of commanded exposure. Howe ver, the autopilot does not provide access to synchronous inertial information, so integration is not an option. Only a two dimensional position correction factor, is considered in the georefer encing solution. The third row of 5-1 is fixed because the altit ude is assumed constant due to th e closed-loop altitude control. The correction factor is similar to dead reckoning, where groundspeed ( ) and heading ( ) are extrapolated for some time delay to predict a future position. Equation 5-1 shows the 2-D position offset of the camera after time. In this case, an 87 millisecond time delay with a ground speed of 14 meters/second would result in a 1.22 meter world position change in the heading of travel. errorworld syncrsin() cos() 11sync errorworldEast North r (5-1) 55

PAGE 56

Figure 5-1. Atmega 128 MCU firmware flowchart Figure 5-2. Virtual Cockp it software screen shot 56

PAGE 57

57 Figure 5-3. PolarisLink software Figure 5-4. Camera control window

PAGE 58

CHAPTER 6 CAMERA CALIBRATION AND INITIALIZATION Camera Calibration is necessary in order to improve the accuracy of the projective geometry solution for geo-registration. A m odel of the cameras imaging sensor and optics system as well as the orientation of the sensor at time of exposure is requi red for accurate direct georeferencing. This process is called solving for the interior a nd exterior camera parameters. Both of these calibrations are typically done in a lab setting; however, fo r this application it is not always possible to do so. A fast method was developed for estimating these parameters by using a modified camera calibration toolbox for Matlab. Computing the exterior orientation parameters in order to isolate the boresight of th e camera refers to the camera initialization and is usually performed at the flying site before a survey mission. Calibration Software Well documented, camera calibration software for Matlab exists that uses images of a known grid size to extract both interior and exteri or camera parameters [8]. Functions of this toolbox were modified for automa tion and field initialization. The interior calibration parameters include th e focal length, scale factor, principle point offset and lens distortion coefficients. A calibra tion procedure in [8] is used to determine these parameters. Since these parameters are intrinsi c to each payload they should only have to be calibrated once. These values are stored in a file called Calib_Results.mat and are saved on the SD card where the UAV state values are written to. Later software assumes that the calibration parameters on this SD card match that of the payload camera. The complete camera model used here includes radial and tangential distortions and can be seen in Equation 6-1. This equation is implemented in the resampling process to undistort th e images and to improve the accuracy of the georeferencing solution [33]. In Equations 6-2 through 6-5, and represent iu iv 58

PAGE 59

the pixel locations in the image plane of the object coordinates (i x ,i, ) and a camera focal length,yiz f Parameters and () r iu() r iv represent the change in pixe l location in the image plane from a radial distortion contribution. Terms represent coefficients of radial distortion of the lens and is only a substitution term representing th e radial distance of pixel i from the principal point. In this case, a fourth orde r estimation for the radial distortion is used. Parameters 1k2kir()t iu and ()t iv represent pixel offsets in the image plane due to tangential d Terms 1p, 2p are coefficients of tangen tial distortion. Parameters iuand iv reprsent te location of object points (iistortions. ehactual x ,iy,iz) in the image plane. Constants uD and vD are convion tors that are necessary to tran sform metric units to pixels. Th ese are typically taken from image sensor datasheets. The term uers fac the s is a scale factor that is adju sted in the iterative solution to solve for lens distortions unknowns. Typically, u s is close to 1. Terms and 0 are the 0u vcoordinates of the im age center, al so known as the principal point. ii iv zi iux f y 2 2kii ii (6-1) (6-2) (6-3) ...) ...)i ir rr r () () r r () () 4 12 42iii iiiuukrk vvrk 1( ( 2 2 u vi i 2 12 2 212( 2(i iupuvp vpuvp 2 2 ) )i i t t 59

PAGE 60

()() 0 ()() 0() ()rt i uuiii rt i viiiuu Dsuuu vv Dvvv (6-4) 2 iiruv2 i (6-5) The following describes the process used to perf orm a self-calibration to determine interior camera model parameters. For calibrating the Canon A650 sensor in the Polaris, the camera was left in the protective carbon box a nd installed into the fuselage; the wings were left off and a battery was installed. A grid of black and white squares was prin ted out and mounted to a flat sheet of Plexiglas; taught as possible. The squares we re measured with a micrometer and determined to be 28.45 mm per side. Using the wireless video transmission and the PolarisLink camera control window, a series of eleven pictures were taken at varying he ights and orientations about the calibration grid. The images were ta ken at a 640 x 480 resolution because the camera calibration toolbox has to read in all the images in order to start the calibration process; this is very RAM intensive. The reduced resolution ha s to be taken into account for later for undistorting high resolution images. Images were tr ansferred to a computer working directory for calibration via the PolarisView GUI The calibration photos can be seen in Figure 6-1. (It should be noted that near the edges there is photo occlusion from the UAV fuselage.) The software uses grid corner intersection detection for control points. Internal intersections of other grid point s can automatically be determined if the four corners of the enclosed box is selected. The software can also count the total number of squares in the picture, but if the software guess is wrong, then the user can enter it manually. The order that the user clicks is important, therefore numbers were added to the actual calibration grid so that in the calibration software, the corner numbers are visibl e and provide a guide in which corner to start 60

PAGE 61

with. The corner intersection must start internal to the grid, not on the outer most squares. The first selected corner will dictate the grids origin, the second poin t establishes a line connecting the first and second points. This is the Y-axis of the photo. All four corners should be selected in clockwise fashion. The righthand rule dictates the Z-axis of the checkerboard. The software produces predicted edge corners for the center pi xels. If they are satisfactory, the user can continue, else an initial radial distortion factor is asked to be input. In the case of the Canon A650, an initial estimate of -0.25 was required in order to get central edge detection to appear correct (Figure 6-2). The size of the actual squares is input next. After all images are brought in through this fashion, the initial calibration can commence. There are two procedures used to calculate interior parameters here. The first is a closed form solution that is not based on any lens distortions. The second procedure involves a non-linear op timization that minimizes the reprojection errors on to the photo, namely 9 degrees of freedom in the intrinsic parameters [33]. The radial and tangential effects of the lens ca n be seen in Figure 6-3. A complete distortion model and interior calibration parameters including uncertainties can be seen in Figure 6-4. The complete model combines the e ffects of tangential and radial distortion in a single plot, effectively showing true pixel lo cation due to lens distortion. Th e coefficients shown in Figure 6-4 can be substituted in to Equation 6-4 to re alize the camera model for the Canon A650. It is obvious that this particular camera suffers from a decentering error, mean ing that the effective center of the photo is different from the geometric center [33]. Now that the interior parameters are realized, images can be resampled in order to ta ke into account lens dist ortions (Figure 6-5). The software also has the ability to solv e for exterior camera parameters. These parameters give the relative location of the camer a versus the grid and consist of a translation vector and a rotation matrix. The software uses the grid corner locations relative to each other to 61

PAGE 62

determine the roll, pitch and yaw of the camera at the time of exposure. Figure 6-6 shows the orientation of the UAV fuselage when the calib ration images were taken. However, these orientations are meaningless for the moment. The same procedure to find the exterior parameters is used in field initialization, only the results will give th e boresight or relative orientation of the camera versus the IMU. Initialization Jig A method for determining the pa yloads boresight at the fiel d was developed to calibrate the exterior parameters quickly and easily before flight. Theoreti cally, if the camera was never remounted before the previous flight, this pr ocedure could be skipped to save time. Since direct georeferencing will be accomplis hed using only the measurements from the UAVs INS, it is necessary that the boresigh t misalignment be repr esented relative the measurements of this INS. The Polaris requires an inertial initialization procedure in which the trimmed flying condition is spec ified to the INS by holding the ai rcraft level. This level calibrated condition is what the autopilots closed-loop control tries to regulate about during straight flight paths. Figure 4-6 shows these control loops. The concept of estimating the exterior orientation parameters consists of init ializing the autopilot and then taking a picture of the calibration grid. A bubble level was used on the aircraft and on the calibration grid to ensure that both platforms were orthogonal to each other. The resulting exterior parameters represent purely the boresight misalignment of the camera (ima ge frame) relative to the IMU. A special initialization jig was designed to facilitate this procedure easily. The initialization jig consists of a heavy-duty fluid head tri-pod with a fuselage cradle mounted on top. The cradle has a hole cut out of the bottom for the payload camera to view through. The cradle is lined with a mid-density foam to secure the fuselage in place. Once the plane is placed in the cradle, a 2-axis bubble level is placed on a flat portion of the fuselage. The 62

PAGE 63

tripod is manipulated as to leve l the aircraft indicated by the bubbl e level. The tri-pod is then rotated to north using a hand held compass. Using the wireless video from the payload camera, the calibration grid is then placed below the payload hole enough to view most of the grid in the fiel d of view of the camera. A platform of any type can be placed underneath the Plexiglas calibration grid apparatus. The grid is then leveled and twisted toward nor th using the 2-axis bubble level and compass respectively. Figure 6-7 show s the initialization jig assemb ly and 2-axis bubble leveling. The Polaris Link software is used to run the in itialization procedure. First, the Polaris main power is turned on and left on for about 5 mi nutes as to let the avionics come to a steady state temperature. Then, the Initialize Sensors button is used to initia lize the INS to a level flying condition. Once all axes of the aircraft states read zero on the Virtual Cockpit, the GPS Sync Shutter button can be pressed. After a few second delay due to the memory card writer initializing, a picture will be taken of the calib ration grid. The wireless video will confirm this step. The GPS Sync Shutter can now be turn ed off and is ready for a survey mission. The Polaris now has the data needed to calculate th e boresight misalignment for direct georeferencing during post-processing. 63

PAGE 64

Figure 6-1. Images used for camera calibration. Original images are in color, but calibration software resamples to black and white. Figure 6-2. Internal grid in tersection points are based on square sizes and initial lens distortion coefficient kc. 64

PAGE 65

A B Figure 6-3. Lens distortion m odel of Canon A650. A)Tangential lens distortion. B) Radial lens distortion Figure 6-4. Complete lens dist ortion model. Radial and tangential effects have been combined into a complete model. 65

PAGE 66

A B Figure 6-5. Comparison of distor ted picture based on lens parame ters. A) Original distorted picture B) Undistorted picture Figure 6-6. Exterior camera para meters for indoor calibration. 66

PAGE 67

67 A B Figure 6-7. Field initializati on process for determining camera boresight misalignment A) Initialization Jig used for extracting exte rior calibration parameters. B) Bubble leveling technique used to ensure or thogonality between ca libration grid and UAV.

PAGE 68

CHAPTER 7 RAPID MAPPING Synchronized imagery, aircraft orientation a nd camera calibration da ta are stored on two independent SD cards aboard the Polaris UAV. A picture number on both cards is the common link between both the set of imag es on one card and the correspondi ng data on the other. Rapid mapping of the imagery collected from the Polaris is the primary goal of this system. Often, the UAV either flies over terrain that is impossible to set out grou nd control points on or has too many similar features to rely on an auto-det ection algorithm to genera te control points. Therefore, a direct georeferencing method was used to compute mapping coordinates and georegister the images. There was no COTS softwa re that existed that could handle the output format of the Polaris in a rapid batch fashion, so specialized software wa s developed to promote user-friendly, autonomous processing. The software was developed in Matlab and written general enough to be easily ad apted to future UAV platforms. Coordinate Frames In order to properly locate a target on the ground or realize photographic capture boundaries, several coordinate frames must be esta blished. There were three coordinate frames necessary in the process of overlaying geo-registered images into Google Earth, namely the body frame, image frame and the world frame. The image frame was chosen to be the computational frame and is where all aircraft c oordinate frames are transformed to for spatial analysis and direct georeferencing. The worl d frame is where projected ground coordinates exist. All axes naming conventions were assigned arbitrarily in a right-handed fashion in order to resolve any inconsis tencies in motion axes amongst aeronau tical engineering, geodetic science and traditional photogrammetry. 68

PAGE 69

The world frame is defined with respect to th e reference ellipsoid. The reference ellipsoid is a simplified model of the shape of the earth [39]. It can be e xpressed by simple equations in which a typical ellipsoid is flattened at the poles; there exists a semi-major axis (a) and a semiminor axis (b) [39]. The GPS receiver used on the Polaris UAV uses the WGS84 datum values for (a) and (b) [40] (Table 7-1). The z-axis of the world frame is orthogonal to the reference ellipsoid, the y-axis is always a ligned with geodetic no rth and the x-axis is a cross product of the y and z axes to form a right-handed coordinate sy stem [39]. In other words, the world frame is tangent to the reference ellipsoid; the x-axis points to the east, the y-axis points toward north and the z-axis is orthogonal to the earths surface (also known as East -North-Up system). There are two instances of the world frame used here and are identical in a 2-D sense; the first is fixed to the ground and the other is a distance h off the ground, centered at Polaris GPS receiver. A position vector in the world frame attach ed to the UAV is shown in Equation 7-1. UAVworld worldworldx y h r (7-1) The height h is the combination of the UAVs altitude, alt measured by the barometric (static) pressure sensor on the autopilot and the difference in spacing between the autopilot and GPS receiver. The GPS output altitude was not used in the direct georeferencing solution. The autopilot in the Polaris UAV uses geodetic latitude ( ) and longitude ( ) for 2-D navigation (Figure 7-1). However, it was ne cessary for the location of the airc raft to be represented in units that were consistent amongst other coordinate systems. That is, all units of spatial measure are represented in meters. Latitude and longitude da ta from the aircraft was therefore converted to UTM coordinates during post-processing. The Po larisView software uses a UTM (Universal 69

PAGE 70

Transverse Mecator) Cartesian coordinate system that complements the make-up of a digital image. The body frame is an orthogonal frame whose origin and orientation is arbitrary [39]. In this case, the body frames origin is fixed to th e center of the IMU (inert ial measurement unit) on the autopilot and is where the phys ical inertial measurements are made. The location of the IMU was basically on the center-of-mass of the aircraft. The coordina te system within the body frame was defined where the x-axis is ou t the right wing, the y-axis is out the nose and the z-axis is out the top of the UAV (Figure 7-1). A position vector in the body frame is given in 7-2. body bodybody bodyx y z r (7-2) Any rotation of the body frame axes relative to the initialized level condition with respect to the world frame describes the aircrafts at titude. The angles by which the attitude is comprised are called Euler angles [41]. The pitch ( ) angle of the UAV is determined by the amount of rotation about the x-axis where a pos itive pitch describes a nose up condition. The roll ( ) angle is the measure of rotation about th e y-axis where a right wing down condition describes a positive roll. The yaw ( ) angle is a measure of rotation about the z-axis where a positive yaw rotates the nose in a counter-clockwi se direction. However, heading used for navigation is measured in a clockwise fashion; therefore a sign change was implemented in the PolarisView software. The image frame is a coordinate frame in which the perspective center, P.C. is the origin. The naming convention of the axes stayed consistent with the other previous frames to avoid intermediate transformations amongst coordinate fram es. The x-axis is defined out of the top of 70

PAGE 71

the camera, the y-axis is defined out of the left side of the came ra and the z-axis is defined out the opposite direction of the lens (Figure 7-1 & Figure 7-2). A position vector in the image frame can be seen in 7-3. image imageimage imagex y z r (7-3) The world frame fixed to the ground is where resulting direct georeferenced coordinates lie. The estimated ground coordinates dictated ho w the image was to be geo-registered. The world frame fixed to the ground is also expressed in UTMs. Polari s 2-D position at any point in time is common among both instances of the world fr ame. The origin of the UTM projection lies at the central meridian and the equators inte rsection [42]. A position vector in the mapping frame is shown in 7-4. The world x coordinate expresses distance east of the origin and the coordinate expresses distances north of the origin while the worldyworld z coordinate expresses the distance above the reference ellipsoid. For fl at terrain testing that was conducted here, world z was fixed to zero. groundworld worldworld worldx y z r (7-4) Direct Georeferencing Traditional image-to-ground georeferencing st rategies involve th e inclusion of ground control points that have to be manually determ ined from land based GPS or similar spatial measuring tool. Often, areas in which UAVs are operated have indistinguishable ground 71

PAGE 72

features, eliminating the possibility of using ex isting objects as GCPs. Th erefore, in order to have ground control in images taken by the Polaris, manual ground-based objects would have to be laid. Small footprint photographs taken by th e Polaris would require a large number of these ground control points spread. This process would be very time and labor consuming and defies the quick-to-fly ability of the Polaris. Consequently, the concep t of direct georeferencing was implemented to estimate such ground points. Due to the proprietary nature of the autopilot and wireless bandwidth constraints, the direct georeferenced solution is based solely on the discrete aircraft state parameters at the time of exposure. The process of direct georeferencing tran sforms image coordinates into ground world coordinates through the location and orientation of the camera at the time of exposure. Using the developed camera calibration process, synchroni zation error, and GPS/ strap-down INS, these parameters can be computed. The mathematical expression for the cameras positioning can be modeled as a spherical joint r obot with a three degree-of-freedom end-effecter, much like mechanisms described in [43]. In order to prop erly assign the position and orientation parameters of the camera for direct georeferencing, the world frame attached to the UAV and the body frame had to be transformed into the image fram e. This was accomplished by a series of rigid body rotations, and link offsets. The IMU was treated as the UAVs global origi n. Distances to the image frame origin (perspective center, P.C.) and GPS receiver were measured relative to the IMU. In addition, the boresight of the camera had to be included as a ro tational bias that impacted aircraft Euler angle measurements with respect to the image frame. The variable was introduced to represent the spatial difference between the GPS receiver and the IMU in body frame coordinates (7-6). The bodyGPSr..body P Cr parameter describes the location of the camera versus the IMU (7-7). The Euler 72

PAGE 73

angles of the aircraft were expr essed as a DCM (direction cosine matrix) and can be seen in 7-8. The y x symbol represents a 3x3 rotation matrix whic h relates the orientati on of one coordinate frame to another. The subscript x represents the original coordinate frame and y represents the subsequent coordinate frame. The rotation matrix was constructed using methods outlined in [43]. In 7-8, an s was used in place of sine and a c was used in place of cosine. All sequence of rotations used for photogrammetric an alysis were X,Y,Z which corresponds to pitch ( ), roll ( ) and yaw ( ) in Euler angles. The 3-D estimate of the location of the image plane origin in world coordinates measured from th e body frame origin can be seen in 7-9. world bodyGPSbodyzody body bodyhalt (7-5) GPSb bodyx y z r (7-6) body body body .. body PCx y z r (7-7) c c world bodycc ss csssccssssc s scsccsscc sc world image (7-8) .. ..image bodyworld worldworld bodybody PC GPSsync PCGPS error rrr rr (7-9) 73

PAGE 74

After the 3-D position of the camera was realiz ed, the effective orientation of the camera was determined. In the initial ization section, a method for determining the boresight of the camera was devised. This rotational bias was combin ed with the aircrafts attitude to render the image frames actual orientation relative to the wo rld frame at the time of picture capture. The rotation matrix that describes th e body frame axes relative the image frame (camera boresight) is similar to 7-8, except x represents a rotation a bout the body frame x-axis, y about the y-axis and z about the z-axis (7-10). A matrix multipli cation of the Euler angle DCM and the cameras boresight DCM provided the correct orientat ion of the image plane (7-11). body imagecyczcyszsy cxszsxsyczcxczsxsyszsxcy sxszcxsyczsxczcxsyszcxcy (7-10) *world body image bodyimageworld (7-11) Once the position, .. world P Cr, and orientation, of the camera sensor were established, ray projectio n intersections from the image fr ame to the world frame could be computed. These intersections in world coordinate s are the estimated captured coordinates of the digital image. The assumption of a flat-earth model was made for photogrammetric analysis in the PolarisView software. body imageDirect Linear Transformation Method A photograph contains three dimensional features in a two dimensional space. In order to locate desired features in the photograph, a mathematical mo del describing the relationship between the photo and captured space is concei ved. Many different models for this exist, however, the direct linear transformation (DLT) met hod is selected, here. 74

PAGE 75

The DLT method uses an ideal pinhole model of the camera, where the image plane is offset in the negative z direction a distance, f, known as the focal length, away from the camera sensors origin [32]. A collinear ity condition in analytical photogram metry states that at the time of exposure, any object point on the image plane and its photo im age on the earth all lie along a straight line in three-dimensiona l space [44]. The image plane is devised of pixels in an x,y tabled fashion and has a central pixel location called the principal poin t [1]. A vector originating from the pinhole to each indivi dual pixel in the imag e plane describes the composition of the vector that continues through th e image plane to the ea rths surface (Figure 73). One mathematical expression for the collinea rity equations is shown in 7-12, which describes the vector from the pinhole to a spec ific point on the image plane [7]. 11 ..12 ..13 .. 0 31 ..32 ..33 .. 21 ..22 0()()( ()()( ()(worldworld worldworld worldworld iPCiPCiPC i worldworld worldworld worldworld iPCiPCiPC worldworld world iPCiP imxxmyymzz xxf mxxmyymzz mxxmyy yyf ) ) ..23 .. 31 ..32 ..33 ..)( ()()(world worldworld CiP C worldworld worldworld worldworld iPCiPCiPCmzz mxxmyymzz ) ) (7-12) The terms x i and refer to image plane coor dinates of a desired point in world coordinates. The coordinates y ii0 x and refer to the location of the principle point of the image plane. The0y f parameter is again the focal length of the camera. The x xm.. world PCxterms correspond to the elements of the matrix given in a row/column subscript. The and terms have all been defined previously as origin of th e image frame in world coordinates. Coordinates and are the location of the ground-based object, they represent the estimated ground coordinates of point iin the world frame. Since the PolarisView software assumes the body imagworl ized.. world PCy.. world PCzworld ixworl iyd 75

PAGE 76

ground flat, is equal to zero. Equation 7-13 shows the general form of the direct georeferencing equation where is the estimated world coordinates of an object j from the image frame projection and world frame intersection. The s term is a scale factor that is based on ground level elevation; in this case it is fixed because the flat -earth model is being used. world izworldjrworld j..*worldimageworld imagej PCs rr r (7-13) Photo Geo-Registration At the time of exposure, the UAV is not or thogonal to the ground which affects the image area coverage of the photograph. The geometry of the resulting image undergoes a projective transformation that distorts the actual area co verage and therefore the land area each pixel contains. Therefore, the digital images are resa mpled based on interior camera parameters and the collinearity equations. The resulting physical image will appear distorted; however, the land area that each pixel contains will be approximately equal allowing for the image to be registered into a defined mapping space. The algorithm implemented for resampling the images into a projective space moves the discrete pixel element in the image plane space to a blank world space dictated by the projective transformation. The resampling process employs a bicubic interpolation that results in a smoother color transition and less artifacts over th e nearest-neighbor or bilinear interpolation. The geo-registered image corner points are geor eferenced to a Cartesian Transverse Mercator Projection using the WGS84 datum and UTMs. Blank black pixles are used to keep rectangularity of the output image in order to geo-register image into Google Earth. The corner coordinates are converted to ge odetic latitude and longitude fo r importation into Google Earth via a KML file. Figure 7-4 shows seve ral examples transformations. 76

PAGE 77

PolarisView Graphical User Interface After previous missions, each photograph had to be manually adjusted and registered into Google Earth for analysis. This was very time consuming and took as long as two weeks to attempt to fit the photographs. Software was written to automate the geo-rectification process and can output images to Google Earth as soon as they image is resampled. Various options in software allow filtering of selected images in a data set. Geo-registered images and linking KML files are transferred to a co mmon directory automatically to be viewed in various GIS toolsets. Google Earth was chosen in this case because the so ftware automatically deals with image pyramiding, a scheme by which the resolution of the photos are incr eased or decreased depending on zoom height. Using a simple SD to USB adapter, data can be moved seamlessly to appropriate hierarchal directories for decompiling and mappin g. After the transfer of the original data, PolarisView automatically formats the SD cards and recopies calibration files back to the SD card for another flight. A brief overview of th e software and a typical mapping process can be seen below. The user has to complete several steps in se quence in order to output map-able imagery. First, the cameras relative pos ition and orientation must be de termined. A separate window is launched that allows the user to input locations of the camera and GPS receiver relative to the autopilots IMU so all coordinate frames can be tr ansformed into the camera frame (Figure 7-6). PolarisView uses methods of exterior camera para meter extraction seen in Chapter 6 to calculate the boresight misalignment (Figure 7-7). All intr insic and extrinsic camera parameters are saved to a text file so a recalibration is not necessary the next flight unless the camera payload is removed. The horizontal (cross-track ) and vertical (parallel-track) field of view are calculated and also stored on the memory card. Next, the AS CII represented binary ai rcraft state data in 77

PAGE 78

IEEE754 floating-point format must be parsed an d converted to decimal form. The Parse UAV Data button runs all the necessary functions to handle this. PolarisView generates a text file that contains all converted elements in a tab-delimited format for separate analysis. The user has two options at this point; either start output mapping or plot estimated image area coverage. The latter is recommended first in case there are areas that is not of interest from the flight. The area coverage assessment pl aces translucent yellow polygons over Google Earth rendered maps outlining predicted area covered du ring that particular photograph (Figure 7-8). The user can filter the data by limiting roll a nd pitch angles or by limiting the distance between nadir points of photographs. This separate filtered data is also ma de available in a text file and linked in a KML to the polygon overlays. A checkbox entitled Output to Google Earth Realtime generates immediate polygon overlays or places completed rectified image on Google Earth. Otherwise, the user could launch the KML f ile associated with that data set run. Finally, the Generate Rectified P hotos button starts the or iginal image to digitally rectified directly georeferenced image process. The option to redu ce the resolution is provided to accelerate the mapping process. A status box wa s provided to show completion of the data set. The process can be stopped at anytime by checking the stop box; the process will suspend after the current picture is finished. All overlays KMLs and data text files are stored in the output directory chosen by the user. All rectified imagery and image KMLs are stored in the output directory\Photos directory. 78

PAGE 79

Table 7-1. The WGS84 reference ellipsoid values Parameter Measure Semi-Major Axis (a) 6378137.0 meters Semi-Minor Axis (b) 6356752.3142 meters Eccentricity (e) 0.081819191 Figure 7-1. Definition of coordinate frames used in direct georeferencing 79

PAGE 80

Figure 7-2. Definition of Im age Frame Coordinate System Figure 7-3. Example of DLT projections from camera frame origin through image plane to ground coordinates in mapping frame sa tisfying collinearity conditions. 80

PAGE 81

A B C Figure 7-4. Digitally Geo-regist ered Photos. All photos were taken at approximately 150 meter altitude .A)Original image taken over a levee lock system in South Florida. B) Distortions with 3 degees of roll, -4 de gees of pitch and 5 degrees of yaw. C) Distortions -10 degrees roll, -10 degree pitch and 35 degree yaw. Figure 7-5. PolarisView GUI software package 81

PAGE 82

Figure 7-6. Offset menu in PolarisView for determining camera relative position and orientation Figure 7-7. Example of calibrati on image of taken just before survey flight. Used for determining camera boresight misalignment in PolarisView. 82

PAGE 83

83 Figure 7-8. Example of area coverage assessmen t generated from PolarisView. These yellow overlays represent estimated photogra phed area at time of exposure.

PAGE 84

CHAPTER 8 MAPPING OUTPUT AND ANALYSIS Data Set A flight test was conducted to assess the spatial accuracies of the si ngle photo direct georeferenced solution. Residual comparisons in object locations were a ccessed directly through geo-registered photos. This form of comparis on simulates the real-wor ld application of the Polaris UAV for rapid mapping missions and tests all elements of the georeferencing, and georegistration processes. The compen sated data set and target attri butes used for analysis can be seen in Appendix A (Table A-1 A-2). The controlled surveying fli ght was conducted over an ag ricultural field with ample space to setout ground control points. Disti nguishable numeric targets measuring 1.5 x 1.5 meters were placed in random locations all ov er the field (Figure 8-1). A WAAS enabled handheld GPS was used to take the coordinates of the center of the target and to ensure that the targets were placed at the same elevation on the landscape, eliminating any concerns regarding the test areas topography. Ot her distinguishable features in the photographs that were recognizable in Google Earth were considered for comparison as well. The flight plan resembled a figure-eight type pattern with elongated central legs for straight and level flight portions (Figure 8-2). The ta rgets were scattered in the central region of the flight plan and were captured from vari ous headings depending on the completion of the flight plan. Varying flying hei ghts were conducted throughout the f light to investigate altitude effects in the georeferenced solution. The survey flight was conducted for approxi mately 25 minutes in which time 234 photos were taken. A subset of 34 photos was used fo r comparison. Fifteen of the pictures either contained targets or road intersections that were selected for residual analysis. Multiple targets 84

PAGE 85

were often identified in the same photograph as well as in different photographs, giving multiple estimates of the ground location of the sa me point at different attitudes. Results The goal of the testing conducte d over the pre-surveyed area was to establish a direct georeferenced spatial accuracy metric as a beginn ing point for the UF program. All photos were registered in Google Earth at their full reso lution to enable number s on the targets to be recognizable (Figure 8-3). The cursor in Google Ea rth was used to scroll ov er the targets center and the latitude/longitiude position was recorde d, establishing the residuals analyzed throughout (Figure 8-4). Plots of the targets surveyed position ve rsus the PolarisView mapped target position (Figure 8-5) is used as a gra phical representation of the erro rs present in the georeferenced solution. The blue circle within the plots represents the surv eyed target an d the red xs represent the locations r ecorded from Google Earth. Multiple xs are seen in some plots because multiple images identified the same targ et. A plot of the orientation of the UAV and camera at the time of commanded camera exposur e (Figure 8-6) shows how tilted the UAV was while taking this set of pictures. A complete set of all targets spatial accuracies provide delta east (difference in East position), delta north (d ifference in North position), and a Euclidean distance measure which expresses the magnitude of the residual s which is the straight line distance between the actual and measured target (Figure 8-7). Root-mean -square errors of the Euclidean distances for all targets are compiled to compare accuracies (Figure 8-8). At first glance, RMS errors in positional accuracies seem quite high, however, when the navigation suite, camera synchronization, and sma ll aircraft technologies are considered, large sources of the error can be partially accounted for. It has been shown in literature that in lowdynamic, near-level flight conditio ns, the attitude solution is ade quate, but is mentioned that in 85

PAGE 86

larger banking and pitch situation the accuracy is compromised [30]. No in-situ flight data between the Kestrel IMU and a precision IMU exists so it is hard to quantif y these inaccuracies. In order to compare between RMS positional erro rs and aircraft orientation, a metric for the magnitude of average aircraft orientation pe r target was devised. This comparison shows a general trend that the larger the magnitude of roll and pitch, the larger the error in estimating a targets position (Figure 8-9). It is the authors opinion that th is is derived from the degraded ability for the IMU to accurately measure larger attitude angl es and not from the photogrammetric solution. A plot of theoretical IMU errors shows if both roll and pitch have an error of 1.5 degrees (not unlikel y at higher attitude angles), Euclidean distance wise, the principal point could be as much as 15 meters off (Figure 8-10). Heading inaccuracies is a metric that was not isolated here, but should be considered. In [ 30] 1.1 degree error was realized. This can cause the most spatial inaccu racies where targets are located near the extents of the photo as they have a larger radius from the center of rotation. One parameter of the direct georeferencing th at is not completely accounted for is the camera sync error. Experimentally and thr ough other source, it was determined that an 87 millisecond delay in camera exposure time exists However, there is no way to properly synchronously link this offset timi ng to autopilot rate output. A plot of the UAVs pitch and roll rates during the first 20 p hotos illustrates the responsiveness of the radial and longitudinal axes (Figure 8-11). A plot of these rates integrated over 87 milliseconds at an altitude of 300 meters was shown to estimate part of the errors in the georeferenced solu tion (Figure 8-12). A contribution of approximately 15 meters of error c ould be present in the to tal solution from sync error alone. 86

PAGE 87

The pre-surveying accuracies of the ground targ ets are only as accurate as the GPS they were recorded with. Due to inclement weather, the long proce ss of using more sophisticated GPS equipment had to be bypassed. Consequentl y, a WAAS enabled hand-held GPS receiver was used and has at best a +3 meter accuracy. This could account for at least 3 meters of error. The current UAS GPS receiver is of standard accuracy, 15 meters (2d rms), however it has been shown that a trajectory projection of the stepped GPS samples can dramatically improve the solution. However, this is done in a post-pro cessing environment where data has already been collected [30]. The current system uses realtime GPS data with extrap olated sub-second output, so it is unlikely that the autopilot is outputting this corrected GPS positioning which may result in an error of at least 3 meters. 87

PAGE 88

Table 8-1. Indicates calculated boresi ght misalignments for the data set Roll (Deg) Pitch (Deg) Yaw (Deg) -2.4949 8.4322 3.2641 A B Figure 8-1. Examples of ground control target s used for spatial comparison in the mapping process. Targets are 1.5 meters s quare. A) Target 8 B) Target 17 Figure 8-2. Polaris UAV flight route versus numeric target locations 88

PAGE 89

Figure 8-3. Rectified images for data set are rectified and registered from PolarisView into Google Earths interface Figure 8-4. All pre-surveyed point s were located in registered photos. The coordinates of the center of the target were used for co mparison. Other photographs reveal manmade objects that are identifiable in the original map set. These are used for comparison as well. 89

PAGE 90

A B C D Figure 8-5. Plots A through N represent the pre-su rvey location of the target versus measured values for those targets th rough the photos in Google Earth The O represents the actual target and the X s represent imaged coordi nates. A) Target #1 B) Target #2 C) Target #5 D) Target #6 E) Target #7 F) Target #8 G) Target #9 H) Target #10 I) Target #11 J) Target #13 K) Target #16 L) Target #17 M) North street Intersection N) So uth Street Intersection 90

PAGE 91

E F G H I J Figure 8-5. Continued 91

PAGE 92

K L M N Figure 8-5. Continued 92

PAGE 93

Figure 8-6. UAV and camera attitudes at the time of commanded camera exposure 93

PAGE 94

A B C D Figure 8-7. Plots A through N provide residuals for all registered photos A) Target #1 B) Target #2 C) Target #5 D) Target #6 E) Target #7 F) Target #8 G) Target #9 H) Target #10 I) Target #11 J) Target #13 K) Target #16 L) Target #17 M) North street intersection N) So uth street Intersection 94

PAGE 95

E F G H I J Figure 8-7. Continued 95

PAGE 96

K L M N Figure 8-7. Continued 96

PAGE 97

Figure 8-8. RMS of Euclidean distances for all targets. Ss and ns stand for south street intersection and north street intersection, respectively. Figure 8-9. Averaged combined pitch and roll e ffects on positional accuracy apparent in data. A linear fit shows general tendencies. 97

PAGE 98

Figure 8-10. Simulated principal point error on the ground from IMU inaccuracies in pitch and roll. Figure 8-11. Rates of all UAV states during in itial 20 photograph capture. Demonstrates the high dynamic nature of the UAV. 98

PAGE 99

Figure 8-12. Unaccounted for effects of an 87 millisecond synchronization error with high dynamics present. This error represents errors in the principal point projection only from an altitude of 300 meters. 99

PAGE 100

CHAPTER 9 SUMMARY, CONCLUSTION AND RECO MMENDATIONS FOR FUTURE WORK Summary The purpose of this thesis was to access the design and validity of the initial georeferenced output of the Pola ris UAV. An array of custom ized electronic hardware and software was developed to allow the system to function as an autonomous direct georeferencing mapping platform. By comparing photographed target locations to that of their surveyed location, a measure of actual system performance was determined. A test flight of varying altitudes was conducted in which 234 images were taken. An adequate amount of these photograp hs contained the 14 ground control targets. A subset of 34 images was selected to comprise a data set used for spatial analysis. A total of 62 instances of the control points were captured; 4.9% were identified within a 25 meter radius of their actual location, 36.5% within 50 mete rs, 27.6% within 75 meters, 2 1.3% within 100 meters, and 9.8% within 150 meters. Therefore, a 67.62 meter RMS e rror exists in the dir ect georeferenced image solution across all measurements. Since a benchm ark was established, the results of the mapped images were considered a success. Conclusion It needs to be mentioned that the flight area that this test was conducted was constrictive due to a poor location of the ground operator relative to th e flight area. The longer legs of the flight plan had to be shortened to keep the UAV in sight, making them much smaller then normal survey legs when a chase vehicle is often used. This path shortening didnt allow for the closedloop control on the Polaris to settle adequately before initiating a new turn, re sulting in few nearvertical pictures. Judging by the generalized re lationship in camera attitude versus positional accuracy (Figure 8-9), these nonlevel flight conditions combin ed with non-level camera 100

PAGE 101

mounting, resulted in larger error that what is potential for the system. Further, flight testing was conducted just before a fast m oving frontal weather system whic h is known to have drastic ambient pressure differentials. This could have possibly caused the UAV to have erroneous barometric-based altitude readings; seriously eff ecting the georeferenced so lution. Therefore, it is the authors opinion that the results found here are more than likely, a worse-case scenario. Regardless of the error in positional accuracies, current technology can be very useful in large areas where non-distinguishabl e features exist and high spatial accuracy is not needed. For instance, the Lake Okeechobee leveemonitoring project that UF is involved in with the Corps of Engineers calls for rapid assessment of levee conditions before threat ening weather, i.e. hurricanes, encroach. The goal is to be able to identify possible leaks in the levee with visible spectrum and thermal infrared imagery. The curre nt technology of the Polaris (in the authors opinion) could satisfy these missi on requirements. The Polaris c ould fly along the levee, capture georeferenced imagery with either imaging payload, and rapidly map those images so that civil engineers or an operations staff member coul d send repairmen to id entified problem areas quickly, within a ~68 meter RMS radius. Current methods for leak detection involve driving on top of the levee for some 235 miles. With a rang e of approximately 15 miles, 16 flights could be performed to map the whole lev ee in a fraction of the time while producing organized data that can be used for many other purposes. Recommendations for Future Work Based on initial results, many improvements n eed to be considered to increase spatial accuracies in the mapped images. Improvements n eed to come from two main areas of focus which encompass many ideas for bettering the system. Majority of the improvements require hardware that may or may not be availabl e and only future technology will dictate. 101

PAGE 102

Georeferencing Although the lightest platform is currently in use, an addition of a completely separate IMU and the implementation of a WAAS/DGPS G PS receiver with a PPS signal generator is the quickest way to improve the di rectly georeferenced solution. Though very feature rich in features for navigation and localiz ation of video, the Procerus aut opilot is a proprietary system and therefore restricts th e abilities to alter data output. However, the current autopilot should remain as the navigation contro ller and means for subsystem message generation for the UAV. All things considered, the method developed in this thesis for s ynchronizing the camera to state data of the UAV is the best way. However, if another IMU is used just for the payload, it should be mounted directly above the camera to eliminate the lever-arm offset and to isolate the boresight of the camera. In the opinion of the author, the IMU can remain MEMsbased (for weight) as long as an adequate sampling rate exists from the inertial sensors, so a trajectory solution can be generated in the post-processing steps using better developed state estimate techniques such as Kalman Filtering [39]. This will take longer, but can guarantee better accuracy. Similar to how PolarisView has the ability to adjust output re solution to save time, the me thod by which state estimates are generated could also be toggled in cases wher e speed over accuracy is required. The new WAAS GPS receiver can be shared amongst the autopilot and camera IMU and be synchronized by the PPS signal. A processor should remain in the loop to write all raw sensor data and PPS signal flag transmissions to the removable me mory while simultaneousl y issuing a shutter on reception of the PPS. This will allow synchronized rate information to be provided before and aft the camera trigger flag so an integration of the rate data over the camera exposure time delay can take place rendering a better estimated orientation of the camera at the time of exposure. 102

PAGE 103

A new camera should be selected for aeria l mapping purposes. This camera does not necessarily have to be very e xpensive, but should have a larger image sensor and better optics than the current COTS camera. The camera should de finitely have the ability to change features in real-time. A method of determining the e xposure time delay has to be developed. This method should include some sort of in-situ f eedback mechanism, rather then mathematical model, from the camera to the processor since the CCD array uses capacitive elemental excitation; discharge times be tween different photographed e nvironments vary. Further, feedback from the cameras data storage to the processor needs to be implemented since the image file size is also dynamic based on the phot ographed environment. This causes variable camera buffer consumption and varying data card write times. The instance may exist where the camera is issued a shutter command but cannot phys ically store that picture, putting the system out of sync. Also, with a little more work the pro cessor should continue to get attitude data from the autopilot so that it can determine when to ta ke a photo based on aircraft attitude. This would solely be for reducing the amount of photos taken if storage is an issue. In general, for clarity and sharpness of the photos, it appears that highe r relative groundspeeds due to rotational rates produce fuzzier images. Therefore, it is r ecommended that a camera have as high a shutter speed as possible and that the UAV be flown on a s unny day so a low f-stop setting can be used. On the other hand, minimizing the groundspeed woul d achieve the same effect. A supplemental control design implementation could achieve this. The general procedures taken to directly ge oreference the images ar e not necessarily the cause of errors here. Due to the relatively flat relief found in Florida, only rectification of the photos is fine. However, other methods, namely orthorectification, exist to account for variable relief. In the case of the data set, orthorectification will improve the solution only minimally, but 103

PAGE 104

in other mission areas this method may have to be used. In order to accomplish this practically, a more precise scanning-type altimeter where a sw ath of ground elevations can be determined, is needed. Until Lidar is miniaturized, some sort of range finder device could suffice. The stereo image method with aero triangulation exists, however, more photographs will have to be recorded in order to guarantee enough overlap, possibly caus ing digital storage issues. Small Unmanned Aerial Vehicle Control Two separate issues on UAV cont rol should be considered for the next generation UAS. The first is very quickly obtainable and has been a long time coming and the other is a fresh idea based on results presented in this thesis. In most situations, a lawn mower type survey pattern is desirable to cover all extents of a desired area. A software package needs to auto matically generate a fli ght plan based on the defined outer-most bounds of a survey area. Curre ntly, all intermediate waypoints for turns and transect spacing have to be added manually and is very time consuming. Virtual Cockpit provides a flexible method for adjusting these ma nual points and should still be taken advantage of; therefore, the future software should generate the waypoints as a .fpf (flight plan file) so Virtual Cockpit can import them. Parameters su ch as desired: transect width, radius of dogbone turns, closed-loop settling time, airspeed, al titude, image overlap, and field of view should all be allowable inputs. It is well known that small UAVs are suscep tible to jerky, high dynamic behavior due to wind gust, thermals, etc. After realizing actual rates during synchronous imaging on the Polaris (Figure 8-11), recommendations can be made to minimize this effect. The Polaris currently uses an autopilot that has PID controllers for stability control of the aircraft. The controllers work well in most general cases, however, do not perfor m well when an external disturbance, i.e. wind, is introduced into the system A robustifying term in needed in the control design in order 104

PAGE 105

105 to squash or quickly minimize these disturbanc es. Often though, robust control requires a high energy cost for the actuators and doesnt always respond well to cyclic input. An adaptive controller lacks the ability to squash the impulse nature of a disturbance but can account for unknowns in the plant model and disturbance. A Lyapunov-based control design called robust integral of the sign error (RISE) exists that exhibits the effects of both robust and adaptive control while still guarant eeing asymptotic stability [45]. Due to the slow framing rate of the hobby-type servos used in inexpens ive small UAVs, this type of robust control may be required. In wind gust situations, it is the effect of cross-wind that contributes the most to undesired aircraft attitudes; therefore, it is recommended that a RISE-b ased controller be implemented on the roll axis and engaged during au tonomous synchronized image segments. The idea is that the controller uses the additional cameras IMU to dict ate the roll rate performance of the aircraft, thereby optimizing the axial stability for dire ct georeferencing. The controller can be implemented on the subsystem control processor and have accessibility to the camera IMUs high rate output and the servo output of the Procerus autopilot. An adjustable weighted adder within the processor could inject more servo cont rol based on the output of the RISE controller during high dynamic situations while still having the autopilots reliable navigation algorithms controlling majority of th e flight control system.

PAGE 106

APPENDIX A POLARISVIEW GENE RATED DATASET PolarisView provides more data than just the mapped output and a linked KML file. The software produces a tab delimited text file for the entire missions synch image data in a file called Parsed_Pic_Data.txt. This is automatically generated in the working directory selected by the user. Another file called Compensated_Pic_Data.txt is pr oduced in the output directory (Figure A-1). This file displays compensated data based on boresight misalignment, lever-arm offsets and sync error estimates. Filtered values can be produced from the Compensated_Pic_Data by checking the filtered da ta box in the GUI and changing the tolerance values. The filtered data set is stored in a f ile called Filtered_Pic_Data.txt in the output directory. All images overlaid in Google Earth were inspected for targets and other ground control points. Upon discovery, the latitude and longi tude coordinates were recorded for residual analysis. A table was devised with these coordinates and all other data pertaining to a particular target (Figure A-2). 106

PAGE 107

Table A-1. Compensated_Pic_Data file used for rectification, area assessment and mapping in the results section Pic # Roll (Deg) Pitch (Deg) Heading (Deg) Altitude (Meters) Latitiude (DecDeg) Longitude (DecDeg) Ground Track (Deg) Airspeed (M/s) 3 -17.702057 7.348692 171.999131 302.650646 29.51843654 -82.55319974 156.646661 12.5 4 -9.591348 18.24883 141.069226 301.728292 29.51821166 -82.55298799 123.586996 13.15 5 -0.492459 15.027198 142.870648 301.043439 29.51804924 -82.55267542 127.712293 13.25 6 -3.955692 12.825514 148.9731 301.360062 29.51785994 -82.55241721 130.920856 12.9 7 -35.783686 12.467541 121.56131 299.175375 29.5176786 -82.55212351 106.856629 13.05 8 -15.835625 15.708724 21.875323 302.707527 29.51777679 -82.55139808 24.866368 12.9 9 -30.908914 18.050031 354.74449 299.211441 29.51847561 -82.5512511 1.375099 13.45 10 -33.137967 20.063285 290.522165 296.721063 29.51880934 -82.55138992 300.057997 13 11 0.509901 19.094982 274.281143 298.072045 29.51897826 -82.55175619 285.504869 12.6 12 -11.753432 20.355219 300.845786 299.407955 29.51923453 -82.55213669 309.626392 12.85 13 -9.6316 15.769382 274.593254 299.877959 29.51940177 -82.55248942 285.676757 13.5 14 -6.51899 13.153482 263.127387 302.361301 29.51949319 -82.5529238 273.873826 13.2 15 2.234168 17.322453 283.222693 299.727024 29.51960367 -82.55338227 293.698166 13.9 16 -5.079284 4.144479 276.759658 304.298989 29.5197227 -82.5537864 287.911292 13.5 17 -0.17745 14.548824 272.055169 299.706883 29.51982474 -82.5542062 282.23901 13.7 18 0.78446 12.426085 277.589629 300.692384 29.51991923 -82.55463294 288.140475 13.15 19 -0.844378 10.113675 282.245311 298.842248 29.51992009 -82.55500677 292.666842 13.25 20 -5.564566 16.54997 292.645785 298.551953 29.52024608 -82.55541048 302.521716 13.05 21 -33.111688 5.802998 252.182674 297.637701 29.52034568 -82.55575386 262.242783 13.45 22 -31.423657 24.358676 217.209649 296.914525 29.52016768 -82.55604625 220.244976 12.65 88 -27.893995 8.836426 355.93284 285.323833 29.51921026 -82.55378375 352.025269 13.8 96 -14.974885 9.912005 226.624425 230.668836 29.52025628 -82.55599473 228.724752 13.05 162 -6.357696 9.053145 57.548714 110.8328 29.51924313 -82.55333864 58.728174 11.65 163 -12.431692 10.220439 26.989725 107.505261 29.51943048 -82.55307627 29.621918 10.75 164 9.755034 -3.180418 112.179721 96.254604 29.51964905 -82.55235427 105.939896 10.1 165 8.527806 8.425522 145.008612 93.333823 29.5194665 -82.55197601 134.473195 9.5 166 11.196723 3.78087 189.570826 91.136202 29.51923797 -82.55184489 173.319733 9.4 167 1.757409 6.412629 223.474707 92.650568 29.51892043 -82.55188058 211.593314 10.5 168 -12.192761 -0.333726 233.709978 92.766558 29.5186737 -82.55207996 218.583399 10.2 169 -2.872878 7.760992 234.585145 89.325013 29.51843197 -82.55230827 218.125033 9.3 170 8.119902 13.180888 246.560649 85.033348 29.51822896 -82.55256897 231.761428 9.75 171 8.135832 6.568745 296.295096 87.820784 29.5181822 -82.55256982 290.661489 10.35 172 -14.757748 19.170419 295.86055 85.064699 29.51833325 -82.55314355 291.750109 12.1 173 -11.287602 12.117616 281.968584 90.68542 29.51874499 -82.5539836 275.535404 12.3 174 -0.904886 12.010828 275.334292 90.688861 29.51875532 -82.55428102 267.857769 12.05 175 19.378746 6.46507 291.440263 90.991452 29.51878082 -82.55460093 284.81732 12 176 -0.306252 10.48385 302.831246 84.178375 29.5189062 -82.55459942 300.000702 11.15 177 14.143055 5.200333 313.502077 85.313939 29.51912666 -82.5550863 311.746336 10.65 107

PAGE 108

Table A-2. Pertinent data entr ies for a given target. This data was used throughout the thesis. Target # Latitude Estimate Longitude Estimate East UTM Estimate North UTM Estimate Delta East Delta North Euclidean Distance RMS of Euclidean Dis Roll of Cam Pitch of Cam Alti of Cam 1 29.5196 -82.5534 349450.8255 3266561.7946 51.7765 65.9237 83.8256 80.4227 0.5099 19.0950 298.0720 29.5215 -82.5530 349498.0310 3266771.7617 4.5710 -144.0434 144.1159 80.4227 -11.7534 20.3552 299.4080 29.5198 -82.5535 349443.4541 3266583.2854 59.1479 44.4329 73.9781 80.4227 -9.6316 15.7694 299.8780 29.5204 -82.5531 349479.1755 3266644.7682 23.4265 -17.0499 28.9742 80.4227 -9.5913 18.2488 301.7283 29.5201 -82.5525 349537.8204 3266615.2770 -35.2184 12.4412 37.3513 80.4227 9.7550 -3.1804 96.2546 29.5205 -82.5524 349553.8452 3266653.7465 -51.2432 -26.0283 57.4746 80.4227 -27.8940 8.8364 285.3238 2 29.5195 -82.5530 349487.6178 3266543.7901 45.4975 45.8346 64.5820 69.2431 0.5099 19.0950 298.0720 29.5212 -82.5526 349531.7395 3266733.4038 1.3759 -143.7790 143.7856 69.2431 -11.7534 20.3552 299.4080 29.5196 -82.5531 349484.1012 3266563.5668 49.0142 26.0580 55.5104 69.2431 -9.6316 15.7694 299.8780 29.5201 -82.5526 349527.9442 3266616.1848 5.1711 -26.5600 27.0587 69.2431 -9.5913 18.2488 302.3613 29.5202 -82.5529 349503.9168 3266631.8018 29.1985 -42.1770 51.2977 69.2431 2.2342 17.3225 299.7270 29.5197 -82.5524 349552.0294 3266575.8496 -18.9141 13.7752 23.3987 69.2431 9.7550 -3.1804 96.2546 29.5201 -82.5522 349568.5943 3266611.2083 -35.4790 -21.5836 41.5284 69.2431 -27.8940 8.8364 285.3238 5 29.5196 -82.5543 349362.4759 3266559.3177 8.6824 25.8207 27.2414 58.4514 -9.5913 18.2488 302.3613 29.5205 -82.5547 349326.3750 3266658.0056 44.7833 -72.8672 85.5287 58.4514 2.2342 17.3225 299.7270 29.5198 -82.5535 349440.1275 3266581.0022 -68.9691 4.1362 69.0931 58.4514 -5.0793 4.1445 304.2990 29.5197 -82.5547 349322.0797 3266576.0406 49.0786 9.0978 49.9147 58.4514 -0.1774 14.5488 299.7069 29.5201 -82.5539 349403.2208 3266612.1984 -32.0625 -27.0600 41.9553 58.4514 0.7845 12.4261 300.6924 6 29.5194 -82.5542 349374.6832 3266536.9863 3.9034 23.6678 23.9875 61.6016 -9.5913 18.2488 302.3613 29.5202 -82.5547 349327.7003 3266626.6198 50.8863 -65.9657 83.3120 61.6016 2.2342 17.3225 299.7270 29.5195 -82.5534 349450.6937 3266551.9315 -72.1072 8.7226 72.6328 61.6016 -5.0793 4.1445 304.2990 29.5195 -82.5546 349331.3990 3266547.9841 47.1876 12.6700 48.8590 61.6016 -0.1774 14.5488 299.7069 7 29.5199 -82.5536 349437.5557 3266591.6773 61.6578 72.6637 95.2980 52.9778 0.5099 19.0950 298.0720 29.5201 -82.5536 349429.2927 3266611.7391 69.9208 52.6019 87.4979 52.9778 -9.6316 15.7694 299.8780 29.5208 -82.5522 349568.1666 3266695.3423 -68.9531 -31.0013 75.6017 52.9778 -27.8940 8.8364 285.3238 8 29.5195 -82.5545 349349.9204 3266555.6061 3.7911 29.7655 30.0060 59.3118 -9.5913 18.2488 302.3613 29.5205 -82.5549 349309.0652 3266661.2298 44.6462 -75.8581 88.0212 59.3118 2.2342 17.3225 299.7270 29.5198 -82.5537 349423.0461 3266579.5678 -69.3346 5.8038 69.5771 59.3118 -5.0793 4.1445 304.2990 29.5197 -82.5549 349305.0893 3266574.1619 48.6222 11.2098 49.8976 59.3118 -0.1774 14.5488 299.7069 29.5201 -82.5541 349384.0189 3266611.6791 -30.3075 -26.3075 40.1326 59.3118 0.7845 12.4261 300.6924 108

PAGE 109

109 Table A-2. Continued 9 29.5197 -82.5545 349345.5227 3266574.7296 9.4247 30.5769 31.9965 60.6211 -9.5913 18.2488 302.3613 29.5207 -82.5548 349314.6416 3266686.7595 40.3058 -81.4529 90.8798 60.6211 2.2342 17.3225 299.7270 29.5200 -82.5537 349421.7034 3266602.4190 -66.7560 2.8875 66.8184 60.6211 -5.0793 4.1445 304.2990 29.5199 -82.5549 349304.0510 3266598.0066 50.8964 7.2999 51.4173 60.6211 -0.1774 14.5488 299.7069 29.5203 -82.5541 349386.1998 3266637.0326 -31.2524 -31.7261 44.5338 60.6211 0.7845 12.4261 300.6924 10 29.5192 -82.5538 349414.9782 3266519.9324 1.9942 11.3900 11.5632 56.6792 -9.5913 18.2488 302.3613 29.5199 -82.5543 349360.6599 3266590.2666 56.3124 -58.9443 81.5201 56.6792 2.2342 17.3225 299.7270 29.5193 -82.5529 349497.8284 3266524.3674 -80.8560 6.9549 81.1546 56.6792 -5.0793 4.1445 304.2990 29.5193 -82.5542 349372.2805 3266524.0500 44.6919 7.2723 45.2797 56.6792 -0.1774 14.5488 299.7069 29.5193 -82.5535 349442.3866 3266532.4236 -25.4143 -1.1013 25.4381 56.6792 -6.3577 9.0531 110.8328 11 29.5199 -82.5537 349420.9575 3266590.1256 60.8981 81.0977 101.4171 51.7379 0.5099 19.0950 298.0720 29.5201 -82.5538 349412.0323 3266611.4155 69.8233 59.8078 91.9362 51.7379 -9.6316 15.7694 299.8780 13 29.5197 -82.5535 349439.2469 3266573.1442 53.8993 72.4349 90.2881 66.9809 0.5099 19.0950 298.0720 29.5199 -82.5536 349431.0046 3266594.7575 62.1416 50.8216 80.2771 66.9809 -9.6316 15.7694 299.8780 29.5205 -82.5533 349464.6514 3266660.5908 28.4948 -15.0117 32.2072 66.9809 -9.5913 18.2488 302.3613 29.5202 -82.5525 349537.7013 3266628.1362 -44.5551 17.4429 47.8478 66.9809 9.7550 -3.1804 96.2546 29.5206 -82.5524 349553.5463 3266674.9211 -60.4000 -29.3420 67.1500 66.9809 -27.8940 8.8364 285.3238 16 29.5192 -82.5541 349383.3006 3266521.6859 -4.7141 38.9682 39.2523 59.7989 -9.5913 18.2488 302.3613 29.5200 -82.5547 349328.5863 3266605.8807 50.0003 -45.2266 67.4202 59.7989 2.2342 17.3225 299.7270 29.5193 -82.5533 349457.9964 3266532.5476 -79.4099 28.1065 84.2372 59.7989 -5.0793 4.1445 304.2990 29.5193 -82.5546 349337.8395 3266529.3875 40.7471 31.2666 51.3607 59.7989 -0.1774 14.5488 299.7069 29.5192 -82.5543 349364.5343 3266517.5031 14.0522 43.1510 45.3814 59.7989 -11.2876 12.1176 90.6854 17 29.5193 -82.5531 349481.0555 3266524.2590 18.1580 140.0820 141.2540 94.4614 0.5099 19.0950 298.0720 29.5210 -82.5527 349518.6640 3266712.5187 -19.4506 -48.1777 51.9559 94.4614 -11.7534 20.3552 299.4080 29.5195 -82.5531 349477.2644 3266545.2585 21.9491 119.0825 121.0884 94.4614 -9.6316 15.7694 299.8780 29.5199 -82.5528 349514.0755 3266593.9802 -14.8620 70.3608 71.9133 94.4614 -9.5913 18.2488 302.3613 29.5201 -82.5531 349482.5048 3266618.2327 16.7086 46.1083 49.0424 94.4614 2.2342 17.3225 299.7270 29.5197 -82.5526 349532.7912 3266565.3549 -33.5778 98.9861 104.5261 94.4614 9.7550 -3.1804 96.2546 29.5200 -82.5524 349546.9974 3266598.3067 -47.7839 66.0343 81.5097 94.4614 -27.8940 8.8364 285.3238 s_street 29.5198 -82.5562 349179.4676 3266590.0305 -41.6342 -17.8422 45.2963 46.8666 0.7845 12.4261 300.6924 29.5196 -82.5561 349185.8366 3266566.1144 -48.0033 6.0739 48.3860 46.8666 -0.8444 10.1137 230.6688 n_street 29.5213 -82.5567 349133.9060 3266749.4758 11.8268 -50.7019 52.0630 52.0630 -5.5646 16.5500 298.5519

PAGE 110

APPENDIX B DEVICE PARAMETERS Table B-1. Autopilot Sensor Parameters Sensor Description Value Rate Gyros Dynamic Range 9 / sec Frequency Response 14 Hz Resonant Frequency 0.0318 kHz Resolution 0.1 / LSB Noise Density 22 / sec / H z Bandwidth 9 Hz Accelerometers Dynamic Range 10 gs Frequency Response 0.00150 Hz Resonant Frequency 200 kHz Resolution 300 g / LSB Noise Density 0 g H z rms Barometric Pressure Altitude Resolution 0.249 meters Table B-2. Canon A650 IS Parameters Description Value Sensor Type CCD Effective Pixels 12.1 Megapixels Image Resolution 4000 x 3000 pixels Sensor Size Format 1/1.7 Sensor Physical Size 7.60 mm x 5.70 mm Pixel Size 1.90 m Pixel Ground Coverage @ 100 m 23.57 mm Viewing Angle 50.48 x 39.17 Image Compression Motion JPEG Movie Compression AVI Focal Length 7.4 44.4 mm Maximum Aperture f/2.8 (W) f/4.8 (T) Shutter Speed 15-1/2000 sec. ISO Sensitivity 80/100/200/400/800/1600 Exposure Compensation 2 stops in 1/3-stop increments White Balance Auto, Custom 110

PAGE 111

LIST OF REFERENCES [1] Paine, D.P., Kiser, J.D., Aerial Photography and Image Interpretation, Second Edition, John Wiley and Sons, Inc., Hoboken, New Jersey, 2003. [2] Sasse, D. B., Job-Related Mortality of Wildlife Workers in the United States, 19372000, Wildlife Society Bulletin, Vol. 31, No. 4 (Winter, 2003), pp. 1015-1020. [3] Winslow, Megan V., 2008. Four victims identified in plane crash in western Martin County, TCPalm, March 13th. [4] Jones, P., The Feasibility of Using Sm all Unmanned Aerial Vehicles for Wildlife Research, MS Thesis, The University of Florida, 2003. [5] Lee, K., Development of Unmanned Aerial Vehicle (UAV) for Wildlife Surveillance, MS Thesis, The University of Florida, 2004. [6] Abd-Elrahman, A., Pearlstine, L., Perciv al, F., Development of Pattern Recognition Algorithm for Automatic Bird Detection from Unmanned Aerial Vehicle Imagery, MS Thesis, The University of Florida, 2004. [7] Wilkinson, B., The Design of Georefer encing Techniques for Unmanned Autonomous Aerial Vehicle Video for use with Wildlif e Inventory Surveys: A Case Study of the National Bison Range, Montana, MS Thesis, The University of Florida, 2007. [8] J.-Y. Bouguet. Camera Calibration Toolbox for Matlab. http://www.vision.caltech.edu/bouguetj/calib doc/index.html 2001. [9] Federal Aviation Administration: Unmanned Aircraft Operations in the National Airspace System. Docket No. FAA-2006-25714. Federal Aviation Administration, Washington, DC (2007). [10] United States Government Accountability Offi ce: Unmanned Aircraft Systems, Federal Actions Needed to Ensure Safety and Expand Their Potential Uses within the National Airspace System, Docket No. GAO-08511. Government Accountability Office, Washington, DC (2008). [11] Morris, Stephen J., Miniature Spy Planes: The Next Generation of Flying Robot, Frontiers of Engineering, National Academy Press Washington, DC, 2002. [12] Grasmeyer, J., Keenon, M., Development of the Black Widow Micro Air Vehicle, AIAA Paper No. 2001-0127, January 2001. [13] Albertani, R., Hubner, P., Ifju, P., Lind, R., Wind Tunnel Testing of Micro Air Vehicles at Low Reynolds Numbers, SAE World Conference, SAE Paper No. 2004-01-3090, November 2004. 111

PAGE 112

[14] Claxton, D., Johnson, B., Stanford, B., Sytsma, M., Development of a Composite Bendable-Wing Micro Air Vehicle, 10th Annual International Micro Air Vehicle Competition, Brigham Young Univ ersity, Provo, Utah, May 2006. [15] Albertani, R., Boria, F., Bowman, S., Claxton, D., Crespo, A., Francis, C., Ifju, P., Johnson, B., Lee, K., Morton, M., Sytsma, M., Development of Reliable and Mission Capable Micro Air Vehicles, 9th Annual In ternational Micro Air Vehicle Competition, Konkuk University, Seoul, South Korea, May 2005. [16] Pounds, LaTreva (2006) ARAs Nighthawk: Battlefield in the Palm of your Hand, Unmanned Systems, March 2006, Pg 40. [17] Watts, Adam, Bowman, W., Abd-Elrahman, A, Mohamed, A, Perry, J, Kaddoura, Y, and Lee, K (March 2008), Unmanned Aircraft Sy stems (UASs) for ecological research and natural-resource monitoring, Journal of Ecological Restoration, Vol. 26, No. 1. [18] Tian, L., Xiang, H., Autonomous Aerial Image Georeferencing for an UAV-Based Data Collection Platform Using Integrated Navigation System, 2007 ASABE Annual International Meeting, Minneapol is, Minnesota, ASABE Paper 073046. [19] Zhou1, G., Li, C., Cheng, P., Unmanned Aerial Vehicle (UAV) Real-time Video Registration for Forest Fire Monitoring, Proceedings of IEEE International Conference on Geoscience and Remote Sensing Symposium, Volume 4, 2005, pp 1803-1806. [20] Johnson, L.F., Herwitz, S., Dunagan, S., Lobitz, B., Sullivan, D., Slye, R., Collection of Ultra High Spatial and Spectral Resolution Im age Data over California Vineyards with a Small UAV, Proceedings, Int l Symposium on Remote Sensing of Environment, 2003. [21] Simpson, A., Stombaugh, T ., Wells, L., Jacob, J., Imaging Techniques and Applications for UAVs in Agriculture, 2003 ASAE Annua l International M eeting, Las Vegas, Nevada, ASAE Paper 031105. [22] Herwitz, S.R., Johnson, L.F., Dunagan d, S.E., Higgins, R.G., Sullivan, D.V., Zheng, J., Lobitz, B.M., Leung, J.G., Gallmeyer, B.A., Aoyagi M., Slye, R.E., Brass, J.A., Imaging from an unmanned aerial vehicle: agricultural surveillance and decision support, Computers and Electronics in Agriculture, Volume 44, (2004), pp. 49. [23] Pearlstine, L., Percival, F., Carthy, R., Abd-Elrahman, A., Morris, S. (2001). Development of a Practical Unmanned Aerial Vehicle for Natural Resource Sampling. Proceedings of the 18th Biennial ASPRS Workshop on Color Photography & Videography in Resource Assessment. [24] Caltabiano, D., Muscato, G., Orlando, A., Federico, C., Giudice, G., Guerrieri, S., Architecture of a UAV for volcanic gas sampling, Proceedings of IEEE International Conference on Emerging Technologies and Factory Automation, Volume 1, 2005, pp. 739-744. 112

PAGE 113

[25] NOAA (No Author Listed). NOAA and Partners Conduct First Successful Unmanned Aircraft Hurricane Observation by Flying Through Ophelia. http://www.noaanews.noaa.gov/stories2005/s2508.htm [26] Legat, K., Approximate Direct Ge oreferencing in National Coordinates, ISPRS Journal of Photogrammetry & Remote Sensing, Volume 60, (2006), pp. 239. [27] Ladd, G., Nagchaudhuri, A., Earl, T ., Mitra, M., Bland, G. Rectification, Georeferencing, and Mosaicking of Images Ac quired with Remotely Operated Aerial Platforms, ASPRS 2006 Annual Conference, Reno, Nevada, May 2006. [28] Kumar, P., Singh, V., Reddy, D., Advanced Traveler Information System for Hyderabad City, Proceedings of IEEE International Conference on Intelligent Transportation Systems, Volume 6, 2005, pp. 26-37. [29] Hruska, R., Lancaster, G., Harbour J., Cherry, S., Small UAV-Acquired, HighResolution, Georeferenced Still Imagery, Wildlife Society 12th Annual Conference, September 2005. [30] Perry, J., Mohamed, A., Abd-Elrahman, A., Bowman, S., Kaddoura, Y., Watts, A., Precision Directly Georeferenced Unma nned Aerial Remote Sensing System: Performance Evaluation, Conference Preceedings for ION NTM, 2008. [31] Mohamed, Ahmed, Rob Price (2002) Near the Speed of Flight, Aerial mapping with GPS/INS Direct Geo-referencing, GPS World magazine, March 2002, Vol. 13, No 3. [32] Weng, J., Cohen, P., Herniou M., Camera Calibration with Distortion Models and Accuracy Evaluation, Proceedings of IEEE International Conference on Pattern Analysis and Machine Intelligence, Volume 14, 1992, pp. 965-980. [33] Heikkil, J., Silvn, O., A Four-step Ca mera Calibration Proce dure with Implicit Image Correction, Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition, Volume 17, 1997, pp. 1106-1112. [34] Burnside, C.D., Mapping from Aerial Photographs, Second Edition, Halsted Press a division of John Wiley and Sons Inc., New York, New York, 1985. [35] Hemmleb, M., Wiedemann, A ., Digital Rectification and Ge neration of Orthoimages in Architectural Photogrammetry, CIPA International Symposium, Gteborg, Sweden, Volume 32, 1997, pp.261-267. [36] Procerus Technologies, Kestrel User Guide Kestrel Autopilot (firmware version MA8) & Virtual Cockpit 2.3, Version 1.51, www.procerusuav.com, 2006. 113

PAGE 114

114 [37] ATmega128, ATmega12 8L Summary. Atmel Corporation. (2007), Rev. 2467PAVR 08/07. [38] ATmega8, ATmega8L Summary. Atmel Corporati on. (2007), Rev. 2486SAVR/07. [39] Schwarz, K.P and M. Wei (2000), INS/ GPS integration for Geodetic Applications, Lecture Notes for ENGO 623, Department of Geomatics Engineering, The University of Calgary, Canada. [40] Conversion of Geodetic coordinates to the Local Tangent Plane, Portland State Aerospace Society, Version 2.01 (2007). [41] Klein, V., Morelli, E., Aircraft System Identificat ion, Theory and Practice, The American Institute of Aeronautics and Astr onautics, Inc., Reston, Virginia, 2006. [42] Ressl, C., The Impact of Conformal Map Projections on Direct Georeferencing, International Archives of the Photogrammetry, Remote Sensing and Spatia l Information Sciences, 34 (2002) (Part 3A), pp.283-288. [43] Crane, C., Duffy, J., Kinematic Analysis of Robot Manipulators, Cambridge University Press, 1998. [44] Wolf, P.R., and Dewitt, B.A., 2000. Elements of Photogrammetry with Applications in GIS, McGraw-Hill Science/Engineer ing/Math, New York, N.Y. [45] P. M. Patre, W. Mackunis, C. Makkar, and W. E. Dixon, Asymptotic Tracking for Systems with Structured and Unstructured Uncertainties IEEE Transactions on Control Systems Technology, Vol. 16, No. 2, pp. 373-379, (2008).

PAGE 115

BIOGRAPHICAL SKETCH William Scott Bowman was born in St. Petersburg, FL, and spent most of his life there. Growing up, there was not something that had not been taken apart by Scott. He was born an inventor and has always enjoye d working with his hands on mechanical systems. His father provided great guidance to Scott as a young boy always encouraging him to fix it himself. Scott elected to take three drafting classes and a mechanical design class in high school over the typical PE and basket-weaving othe rs took. Scott became an accomplished trumpetist under the private study of Harry Murphy and guidan ce of Robert Schear. He received several state-level solo performance superior ra tings and the John Phillip Sousa award for musicianship. Throughout high school he excelled in Math and Science and was involved in a local model aviation club, providing opportunities to experiment early with a variety of small aircraft and robotic systems. In late June 2001, Scott began at tending the University of Florida where he declared an undecided engineering major. Scotts interest in all forms of engineering made it difficult to select a specific study. Scott became involved in 2003 with the Micro Air Vehicles research lab at UF, under the guidance of Dr. Peter Ifju. The lab provided a research opportunity in composite materials, aeronautics and micro elec tronics. Scott had the unique opportunity to travel abroad for academic competitions with the members of the lab, winning first place on several occasions. In August 2006, Sc ott received his bachelors de gree in electrical engineering and won a first place award for his senior design project, titled MAV Autopilot. In January 2007, Scott made the decision to pursue a masters degree in mechanical engineering with a focus of study on controls and robotics. Scott became very involved with the UF Wildlife and Corps of Engineers UAV program s for the remainder of his graduate school career. Scott graduated in August 2008 with his Master of Science in mechanical engineering.