Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Inexpensive Aerial Photogrammetry for Studies of Whales and Large Marine Animals

Inexpensive Aerial Photogrammetry for Studies of Whales and Large Marine Animals METHODS published: 15 November 2017 doi: 10.3389/fmars.2017.00366 Inexpensive Aerial Photogrammetry for Studies of Whales and Large Marine Animals 1 2 1 3 Stephen M. Dawson *, M. Hamish Bowman , Eva Leunissen and Pascal Sirguey 1 2 Department of Marine Science, University of Otago, Dunedin, New Zealand, Department of Geology, University of Otago, Dunedin, New Zealand, School of Surveying, University of Otago, Dunedin, New Zealand We describe a simple system enabling accurate measurement of swimming marine mammals and other large vertebrates from low-altitude single-frame photogrammetry via inexpensive modifications to a “prosumer” unmanned aerial vehicle (UAV) equipped with gimballed micro4/3 camera and 25 mm lens. Image scale is established via an independently powered LIDAR/GPS data-logging system recording altitude and GPS location at 1 Hz. Photogrammetric calibration of the camera and lens allowed distortion parameters to be rigorously accounted for during image analysis, via a custom-programmed Graphical User Interface (GUI) running in MATLAB. The datalogger, camera calibration methods and measurement software are adaptable to a wide range Edited by: Alastair Martin Mitri Baylis, of UAV platforms. Mean LIDAR accuracy, measured from 10 bridges 9–39 m above South Atlantic Environmental water, was 99.9%. We conducted 136 flights in New Zealand’s subantarctic Auckland Research Institute, Falkland Islands Islands to measure southern right whales. Mean lengths of 10 individual whales, each Reviewed by: photographed between 7 and 15 times, had CVs (SD/mean) ranging from 0.5 to 1.8% Daniel Paul Costa, University of California, Santa Cruz, (mean = 1.2%). Repeated measurements of a floating reference target showed a United States mean error of c.1%. Our system is relatively inexpensive, easily put together, produces David Peel, Commonwealth Scientific and accurate, repeatable measurements from single vertical images, and hence is applicable Industrial Research Organisation to a wide range of ecological questions in marine and terrestrial habitats. (CSIRO), Australia *Correspondence: Keywords: aerial, photogrammetry, UAV, LIDAR, whale Stephen M. Dawson steve.dawson@otago.ac.nz INTRODUCTION Specialty section: This article was submitted to Photogrammetry is widely used in ecological research, often to provide measurements of animals Marine Megafauna, that are impractical or dangerous to capture (e.g., elephants, Rüther, 1982; sharks, Klimley and a section of the journal Brown, 1983; fin whales, Ratnaswamy and Winn, 1993; coelacanths, Décamps et al., 2016). Frontiers in Marine Science Unmanned Aerial Vehicles (UAVs) are increasingly used in ecological research (see Linchant et al., Received: 28 July 2017 2015, for review). Their use as photogrammetric platforms to measure mobile animals, however, is Accepted: 31 October 2017 a recent and evolving development. Published: 15 November 2017 The central challenge for photogrammetry from UAVs is establishment of scale, which, Citation: for mobile animals, must be achieved instantaneously. The simplest approach is to include Dawson SM, Bowman MH, in the image an object of known size, typically a boat (e.g., Whitehead and Payne, 1981; Leunissen E and Sirguey P (2017) Christiansen et al., 2016), or projected laser dots a known distance apart (e.g., Durban and Inexpensive Aerial Photogrammetry for Parsons, 2006; Rowe and Dawson, 2008). Having a boat in the image involves the potential Studies of Whales and Large Marine for disturbance, and our trials with lasers indicated that the dots may not show up on Animals. Front. Mar. Sci. 4:366. doi: 10.3389/fmars.2017.00366 whales at the ranges required, and raised questions about eye-safety. Alternatively, provided Frontiers in Marine Science | www.frontiersin.org 1 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals that camera separation and orientation are known precisely, a software tool we developed to measure images, and quantify stereo camera systems do not require an independent the accuracy and precision of resulting photogrammetric measurement of range. The extra weight and awkwardness measurements of a floating reference target and living whales. of a stereo camera system, however, are impractical on most A building guide for the datalogger, the code to run it, and our UAVs. If the subject can be assumed to be lying in a horizontal measurement software are provided as Supplementary Material. plane, vertical photography using a single gimballed camera, Our intention is to enable others to apply easily the potential of ideally with a low-distortion lens, and a means of accurately UAV-based aerial photogrammetry to their ecological research. measuring height, is far more practical. There are at least three approaches to measuring height. The METHODS AND RESULTS first uses data logged from a high-precision air pressure sensor Platform Choice (e.g., Durban et al., 2015). Provided there are no substantial The competing needs of light weight, high resolution, dynamic atmospheric changes during the flight, such data can accurately range and low distortion led us to the Micro4/3 system, and record altitude. Carrier-phase GPS is also an option; several to the Olympus 25 mm f1.8 lens. Rather than build a bespoke models are light enough to be carried on a medium-sized UAV, we chose to adapt the DJI Inspire 1 Pro (I1P) quadcopter UAV and can support centimeter accuracy. This approach for single-frame photogrammetry. This UAV has a gimballed relies on having an accurate terrain model, and, if over water, camera (X5, 16 MP, Micro4/3), and supports the Olympus 25 mm correcting for tidal height. LIDAR (Light Detection and Ranging) f1.8 lens. Live video feedback is displayed while in flight via an technology involves illuminating a target with pulsed laser light Apple iOS or Android device running DJI “GO,” which allows and measuring reflected pulses to provide information about adjustment of camera and gimbal settings. We use this particular target range. LIDAR was first applied in meteorology (e.g., UAV platform as an example only; the LIDAR datalogger, camera Goyer and Watson, 1963), but has found a very wide range of calibration process and measurement approach presented below applications since, including recent use in obstacle avoidance are applicable to a wide range of multirotor and fixed wing UAV and navigation in autonomous vehicles (e.g., Schwarz, 2010). platforms. Now that lightweight LIDAR units are available, they provide a simpler, practical solution for instantaneous measurement of altitude in UAVs. The system we describe here uses this LIDAR/GPS Datalogger The I1P has no ports giving access to its GPS data or power. technology. Our purpose is to provide an example by adapting an Therefore, we built an altitude measuring system that was completely independent (Figure 1). This comprised a Lightware off-the shelf “prosumer” quadcopter for specialized aerial photogrammetry, via inexpensive modifications. We provide SF11-C LIDAR, Pololu MinIMU-9 Inertial Measurement Unit, results of testing the accuracy and precision of the LIDAR unit, GlobalSat EM506 GPS module and 3.3 v 8 Mhz ProMicro describe the calibration of the UAV camera and lens, describe Arduino-compatible microcontroller (and MicroSD flash card FIGURE 1 | Block diagram of the datalogger. Text on the connecting lines shows communication protocols. Frontiers in Marine Science | www.frontiersin.org 2 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals FIGURE 2 | Components of the UAV photogrammetric system. breakout). The GPS module is used to timestamp the altitude While mean accuracy across replicated tests is very good, data so they can be matched to image timestamps. When the raw data showed occasional outliers, presumably due to the hovering, multirotor UAVs must tilt into the wind to maintain changing shape of the water surface. To investigate how these position. If the LIDAR is rigidly mounted to the UAV, its might be reduced via applying a median smoothing filter (which measurements of altitude will be biased high (by the cosine of the SF-11 can do internally), we conducted a further test from the the tilt angle). For this reason the datalogger system incorporates Clydevale bridge (9.77 m to water surface). We set the LIDAR to an Inertial Measurement Unit (IMU) which measures UAV tilt. output a moving median of 16 measurements (made at 16 Hz). We wrote code to integrate LIDAR measurements and IMU Both unfiltered and filtered data showed high accuracy (mean data into the output data sentences from the GPS module (GGL error +1.7 cm), but large differences in precision. Unfiltered data and RMC-NMEA protocols). Data are logged (at 1 Hz) by the showed maximum errors of +1.74 m and −1.42 m (n = 2508, SD microcontroller to a Micro SD card as .csv files. Power is supplied = 0.118 m) while the filtered data showed maximum errors of by a separate 7.4 v 500 mAH lithium-polymer battery, with its +0.06 m and −0.03 m (n = 2508, SD = 0.014 m). output regulated to 5 v using a 3 A UBEC. Combined weight of this system is <100 g. Camera Calibration System components are non-invasively mounted to the I1P To achieve rigorous scaling of distances to object size, via double-sided foam tape and gaffer tape. We built carbon- the camera’s interior orientation parameters (IOPs) and lens fiber brackets to support the GPS module and power system. distortion parameters must be known accurately. To avoid slight The latter is mounted at the aircraft nose to offset the weight of changes in focal length due to focussing, the I1P’s camera was components added behind the center of balance (see Figure 2). set to infinity focus. We calibrated the camera and lens by taking Detailed building instructions and the code to run the datalogger multiple aerial images of a field containing 88 40 cm stakes placed are available in Supplementary Material. 2 m apart in a 14 × 20 m grid pattern. A high-contrast target was fixed to the top of each stake. To support scaling, we also placed LIDAR Accuracy within the calibration field three precisely measured windsurfer The specifications of the SF11 LIDAR indicate ±10 cm accuracy masts each with a high-contrast target at each end. The I1P for ranges up to 40 m over moving water. To gain a more detailed was flown around the calibration field, stopping to take oblique evaluation, we mounted the LIDAR/GPS datalogging system on images every 45 degrees in a converging pattern, first at 20 m a wooden board, and clamped it to bridges ranging in height indicated altitude, then at an indicated 30 m and 40 m. This set above water from 9 to 39 m. Raw LIDAR measurements were was supplemented with nine vertical shots as the I1P was lowered recorded at 1 Hz for at least 60 s (n = 60–224). We were assured from 40 to 20 m above the center of the grid (Figure 3). The by the manufacturer that subsequent LIDAR measurements are resulting images were loaded into Australis photogrammetric independent. The Kawarau bridge was measured at two different software (www.Photometrix.com.au) in which stake-top targets heights above water (Table 1). were measured and scale defined. A free-net bundle adjustment Frontiers in Marine Science | www.frontiersin.org 3 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals TABLE 1 | Accuracy of raw SF-11 LIDAR ranges over water. Bridge True height (cm) Mean error (cm) % Error CV (SD/mean) (%) Balclutha 937 −1.4 −0.15 0.76 Clydevale 962.5 −3.0 −0.31 0.58 Horseshoe bend 1037 −0.1 −0.01 0.57 Cromwell 1107.5 −0.7 −0.06 1.75 Bannockburn 1128 −5.4 −0.48 2.43 Clyde 1278 −2.0 −0.16 0.97 Kawarau mining center 1604 −1.0 −0.06 0.72 Roxburgh 2002 −3.1 −0.15 0.64 Arthur’s point 2999 −4.4 −0.15 0.53 Kawarau B 3854 −4.0 −0.10 0.39 Kawarau A 3879 3.5 0.09 0.21 Averages −1.96 −0.14 0.87 True distance to water was measured using a weighted fiberglass tape measure. Outliers were not excluded. Vibration from passing traffic was discernable during tests from the Bannockburn and Cromwell bridges and may have contributed to their larger CVs. TABLE 2 | Calibration characteristics of our DJI X5 camera and Olympus 25 mm f1.8 lens according to the 10-parameter model employed in Australis (Fraser, 1997). Camera/lens parameter Abbrev Value Lens focal length fc 24.851372 mm Principal point on autocolimation PPA 0.203089 mm; −0.087931 mm Radial distortions k1 −9.1303 e-005 k2 8.4284 e-007 k3 −3.7862 e-009 Decentring distortions p1 −3.1598 e-005 p2 2.0922 e-005 Pixel scaling horiz/vertical b1 7.0190 e-004 Axial skew b2 −1.4177 e-004 image, applies the appropriate LIDAR measurement and camera calibration data, and prompts the user to indicate the total length FIGURE 3 | Image network used for camera calibration displayed in Australis of the animal (whale in our case) by creating a series of points subsequently to bundle block adjustment. (mouse clicks) from the tip of the jaw, along the spine, to the fluke notch. To allow for some imprecision, the software fits a smooth curve to these points. The software then offers options to measure other features (see Figure 4). The user is required to input data (Mikhail et al., 2001) was performed to establish the calibration on image quality, and can add notes. These data are written to characteristics of the camera and lens combination (Table 2). an Excel-formatted results file, and the software prompts for the The calibration yielded a hyperfocal length 0.6% shorter than next whale. The software also allows measurement of any other the specified focal length. The lens exhibited moderate barrel image of interest, via manually inputting a LIDAR measurement. distortion only, with maximum radial distortion of −56μm and The MATLAB code is available in the Supplementary Material, so marginal decentering distortion of 4μm at 11 mm radius (corner that others can adapt it for their specific needs. of the Micro4/3 sensor). If uncorrected, the combination of uncertainties on focal length and lens distortion would introduce Accuracy and Repeatability of a 1% error in the size estimate. Photogrammetric Measurements Graphical User Interface To measure accuracy of photogrammetric measurements, we We programmed a Graphical User Interface (GUI) in MATLAB took 65 photographs of a 2.716 m floating reference target, from to incorporate the measured camera calibration parameters, 16 to 30 m altitude, the usual range within which we photograph LIDAR ranges and tilt data into the image measurement process. whales, at the end of whale-measuring flights. On each occasion Using an Excel file listing best images, the GUI opens an the target was photographed in the center of the field of view, Frontiers in Marine Science | www.frontiersin.org 4 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals FIGURE 4 | Screenshot of the GUI measurement screen. The purple line is a curve fitted to user-defined points along the spine to measure total length. Yellow lines indicate where the user should click on the whale’s outline to measure body width at 10% intervals of total length. Green lines represent measurements from eye to eye, rostrum to eye rostrum to blowholes and fluke width. TABLE 3 | Measurements of total length (TL) of photographically identified right australis) for which we had at least seven different measurement- whales. grade images each (Table 3). Images had to be sharp, with the tip of the rostrum and fluke notch clearly visible, and the tail flukes Whale ID n Mean TL (cm) SD (cm) CV (SD/mean) (%) apparently flat (i.e., not drooped). AI16D026 7 1,041 6 0.5 The mean CV of total length measurements was 1.2%, AI16D014 15 1,338 9 0.6 surprisingly precise considering that whales are inherently AI16D025 9 1,285 9 0.7 flexible, there is inevitably some imprecision in tracing the spine, and that rippling of the water surface and refraction AI16D033 15 1,437 18 1.2 AI16D029 8 1,358 18 1.3 can compromise pointing accuracy. This CV implies a 95% confidence interval from a single image of a 14 m whale of AI16D017 9 1,325 18 1.4 13.66–14.33 m. There was no evidence of a relationship between AI16D050 12 1,401 20 1.4 CV and whale size. AI16D013 10 1,365 21 1.5 AI16D065 9 1,316 22 1.7 AI16D012 9 1,285 23 1.8 Deployment Good weather made it possible to fly the I1P on 12 days over ◦ ′ ◦ ′ 22 days in Port Ross (50 32 S, 166 14 E), a harbor at the and in each of the corners. Measurements showed positive bias Northern end of the Auckland Islands (26 July−17 August 2016). when the target was photographed in the bottom left corner and Seven flights were made from a 6.6 m rigid-hulled inflatable, slight negative bias when photographed in the top right corner landing on a 1.6 × 1.6 m platform. This proved challenging in (ANOVA, p = 0.015), suggesting that the UAV’s gimbal may not practice, so the remaining 129 flights were flown from the aft have been exactly vertical. Overall, mean error was 1% (2.7 cm, deck of the Otago University research vessel Polaris II. More SD = 4.9 cm). We made no attempt to correct for this small error recently, we have added “handles” to our UAVs (Figure 2), in measurements of whales. and shifted to hand-catching as our standard practice; this Repeatability of measurements was assessed from 10 makes flying from small boats much more practical. Flights photographically identified southern right whales (Eubaleana were generally 10–12 min in duration. To allow a generous Frontiers in Marine Science | www.frontiersin.org 5 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals safety margin, our protocol was to land with at least 40% sells as a complete system for $US20,000. In the field, the I1P of battery capacity remaining. Flights were made in winds of appeared to fly at least as well as the APH-22 hexacopter, and up to 20 knots, generally over smooth water in the lee of a was much easier to use (R.W. Wallace, APH-22 pilot, pers. shore. All appropriate permissions, including a research permit comm). from New Zealand’s Department of Conservation, were in place. Durban et al. (2015), whose APH-22 hexacopter estimates We saw no evidence of any reaction by the whales to the altitude from logged measurements of air pressure, report UAV. similar mean accuracy in measurements of fixed targets (mean Before safe flight, most UAVs must calibrate their Inertial errors; theirs = 0.7%, ours = 1.0%), but much worse precision Measurement Unit, and compass. Despite launching from (their CV = 4.5%; our CV = 1.8%). Their approach, and moving vessels, the I1P calibrated its IMU successfully on all but that of Christiansen et al. (2016), makes no allowance for one occasion, when the vessel was rolling. We occasionally had any lens/camera distortions. Six individual blue whales they difficulty calibrating the I1P’s compass. This may have been to do measured 4–7 times showed ranges of measurements within 5% with a local magnetic anomaly (strong enough to be marked on of the mean for each individual (Durban et al., 2016). This is the nautical chart of Port Ross), or to the substantial steel gantry a smaller sample (of whales and repeat measurements) than on the aft deck of Polaris II. Relatively close proximity (<2 m) we present above (Table 3), but suggests similar precision in to this was unavoidable. On the few occasions when reattempting actual use. This may be because precision is driven more by the calibration or moving the UAV a meter or so was not effective, the flexibility of whales than by errors in the photogrammetric we successfully calibrated the I1P’s compass on the wheelhouse process. Dawson et al. (1995) reached a very similar conclusion roof, away from magnetic materials. The Polaris II is a wooden comparing different boat-based methods of measuring sperm vessel. whales. We did notice an unexpected feature of the I1P’s image The largest contributor to photogrammetric error in our timestamps. These are synchronized from the tablet running system is LIDAR inaccuracy. Ranges measured from stable road “GO,” the camera control software. In remote areas, away from bridges showed occasional outliers. With the whale data, we access to a Network Time Protocol server, the tablet clock minimized outliers by applying a 5 s median filter to the LIDAR is likely to drift. We measured and corrected for this using measurements, centered on the time when the photograph was GPS timestamps of LIDAR data gathered during videoed take- taken. This is a reasonable because while photographing a whale, offs or landings. Time drift on our iPad Air was 9 s over the we seldom change throttle settings (therefore altitude). The 17 days on which we flew. In future remote trips we will consistency of measurements of known whales shows this was daily photograph the time screen of a handheld GPS with the effective. The LIDAR can be programmed to apply a median iPad. filter to raw measurements; enabling this feature increased the We also took along an APH-22 hexacopter equipped with an precision of altitude measurements by more than an order of Olympus E-PM2 and 25 mm f1.8 lens (Durban et al., 2015) and magnitude. two Swellpro Splashdrones equipped with Canon D30 waterproof This system was designed for measuring whales, but could point-and-shoot cameras programmed to take images every 2 s be applied to measuring any large animal swimming at the (Christiansen et al., 2016). The APH-22 flew well and collected surface (e.g., sharks, dolphins) or hauled out on a beach (e.g., data for 2 days, but crashed into the water when the velcro strap pinnipeds), or to measuring the spacing between individuals. It holding its battery in place failed. This incident was in contrast could also be used to measure habitat areas at small scales (e.g., to the usual reliability of this aircraft (e.g., Durban et al., 2015). coral patch reefs). The calculations assume that the object to be One of the Splashdrones flew briefly, and crashed into the water. measured is essentially on a flat plane whose range from the Thereafter neither could calibrate its flight controller, and hence camera is measured by the lidar. Without knowing exactly where did not fly. The local magnetic anomaly may have been the cause the lidar was measuring to, the system would not be suitable for of this problem. measurements of terrestrial animals on uneven ground. ETHICS STATEMENT DISCUSSION The research was conducted under permit 50094-MAR from The purpose of this contribution is to describe modifications Department of Conservation New Zealand. we have made to an off-the-shelf UAV to make it suitable for low-altitude aerial photogrammetry and provide sufficient information for others to be able to replicate it. These AUTHOR CONTRIBUTIONS modifications could be applied to a wide variety of mid- size UAVs. Excluding our development time, the cost in New SD developed the central ideas, conducted all fieldwork, analyzed Zealand of our I1P-based system, including 25 mm lens, iPad the data and led the writing process. MB designed and Air, 6 TB48 batteries and charger was approximately US$5,300. programmed the datalogger. EL developed the measurement The parts to build our GPS/LIDAR datalogging system total software. PS conducted the photogrammetric calibration. All approximately $US340. If you wish to use the GUI described authors contributed critically to the drafts and gave final approval above, you may need to buy MATLAB. The APH-22 hexacopter for publication. Frontiers in Marine Science | www.frontiersin.org 6 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals ACKNOWLEDGMENTS ltd) facilitated a discounted Inspire 1 Pro. Alastair Neaves, Mike Denham and Ray McLennan advised on measuring This research was principally funded by the New Zealand LIDAR accuracy. We thank Chris Fitzpatrick for administrative Antarctic Research institute, with additional support from assistance. Otago University and NZ Whale and Dolphin Trust. SD is especially grateful to Colin Gans for UAV advice. Mike SUPPLEMENTARY MATERIAL Paulin helped with our prototype LIDAR-datalogger. We thank The Supplementary Material for this article can be found online the crew and expedition members on board Polaris II for the 2016 Winter season. Richard Greasley (Lacklands NZ at: https://github.com/EvaLeunissen/Whalength REFERENCES Linchant, J., Lisein, J., Semeki, J., Lejeune, P., and Vermeulen, C. (2015). Are unmanned aircraft systems (UASs) the future of wildlife monitoring? Christiansen, F., Dujon, A. M., Sprogis, K. R., Arnould, J. P. Y., and Bejder, A review of accomplishments and challenges. Mammal Rev. 45, 239–252. L. (2016). Non-invasive unmanned aerial vehicle provides estimates of the doi: 10.1111/mam.12046 energetic cost of reproduction in humpback whales. Ecosphere 7:e01468. Mikhail, E. M., Bethel, J. S., and McGlone, J. C. (2001). Introduction to Modern doi: 10.1002/ecs2.1468 Photogrammetry. New York, NY: John Wiley & Sons. Dawson, S., Chessum, C. J., Hunt, P., and Slooten, E. (1995). An inexpensive, Ratnaswamy, M. J., and Winn, H. E. (1993). Photogrammetric estimates of stereo-photographic technique to measure sperm whales from small boats. Rep. allometry and calf production in fin whales, Balaenoptera physalus. J. Mammal. Int. Whal. Commn. 45, 431–436. 74, 323–330. doi: 10.2307/1382387 Décamps, T., Herrel, A., Ballesta, L., Holon, F., Rauby, T., Gentil, Y., et al. Rowe, L. E., and Dawson, S. M. (2008). Laser photogrammetry to determine dorsal (2016). The third dimension: a novel set-up for filming coelacanths in their fin size in a population of bottlenose dolphins from Doubtful Sound, New natural environment. Methods Ecol. Evol. 8, 322–328. doi: 10.1111/2041-210X. Zealand. Aust. J. Zool. 56, 239–248. doi: 10.1071/ZO08051 12671 Rüther, H. (1982). Wildlife stereo photogrammetry at close range. Int. Arch. Durban, J. W., Fearnbach, H., Barrett-Lennard, L. G., Perryman, W. L., Photogramm. 24, 422–432. and Leroi, D. J. (2015). Photogrammetry of killer whales using a small Schwarz, B. (2010). LIDAR: mapping the world in 3D. Nat. Photon. 4, 429–430. hexacopter launched at sea 1. J. Unmanned Veh. Syst. 3, 131–135. doi: 10.1038/nphoton.2010.148 doi: 10.1139/juvs-2015-0020 Whitehead, H., and Payne, R. (1981). New Techniques for Measuring Whales from Durban, J. W., Moore, M. J., Chiang, G., Hickmott, L. S., Bocconcelli, A., Howes, the Air. US Marine Mammal Commission Report MMC-76/22. Washington, G., et al. (2016). Photogrammetry of blue whales with an unmanned hexacopter. DC, 36. Mar. Mammal Sci. 32, 1510–1515. doi: 10.1111/mms.12328 Durban, J. W., and Parsons, K. M. (2006). Laser-metrics of free ranging Conflict of Interest Statement: The authors declare that the research was killer whales. Mar. Mammal Sci. 22, 735–743. doi: 10.1111/j.1748-7692.2006. conducted in the absence of any commercial or financial relationships that could 00068.x be construed as a potential conflict of interest. Fraser, C. S. (1997). Digital camera self-calibration. ISPRS J. Photogramm. Remote Sens. 52, 149–159. doi: 10.1016/S0924-2716(97)00005-1 Copyright © 2017 Dawson, Bowman, Leunissen and Sirguey. This is an open-access Goyer, G. G., and Watson, R. (1963). The laser and its application to meteorology. article distributed under the terms of the Creative Commons Attribution License (CC Bull. Am. Meterol. Soc. 44, 564–575. BY). The use, distribution or reproduction in other forums is permitted, provided the Klimley, A. P., and Brown, S. T. (1983). Stereophotography for the field biologist: original author(s) or licensor are credited and that the original publication in this measurement of lengths and three-dimensional positions of free-swimming journal is cited, in accordance with accepted academic practice. No use, distribution sharks. Mar. Biol. 74, 175–185. doi: 10.1007/BF00413921 or reproduction is permitted which does not comply with these terms. Frontiers in Marine Science | www.frontiersin.org 7 November 2017 | Volume 4 | Article 366 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Frontiers in Marine Science Unpaywall

Inexpensive Aerial Photogrammetry for Studies of Whales and Large Marine Animals

Frontiers in Marine ScienceNov 15, 2017

Loading next page...
 
/lp/unpaywall/inexpensive-aerial-photogrammetry-for-studies-of-whales-and-large-5X9PcSJ6R6

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Unpaywall
ISSN
2296-7745
DOI
10.3389/fmars.2017.00366
Publisher site
See Article on Publisher Site

Abstract

METHODS published: 15 November 2017 doi: 10.3389/fmars.2017.00366 Inexpensive Aerial Photogrammetry for Studies of Whales and Large Marine Animals 1 2 1 3 Stephen M. Dawson *, M. Hamish Bowman , Eva Leunissen and Pascal Sirguey 1 2 Department of Marine Science, University of Otago, Dunedin, New Zealand, Department of Geology, University of Otago, Dunedin, New Zealand, School of Surveying, University of Otago, Dunedin, New Zealand We describe a simple system enabling accurate measurement of swimming marine mammals and other large vertebrates from low-altitude single-frame photogrammetry via inexpensive modifications to a “prosumer” unmanned aerial vehicle (UAV) equipped with gimballed micro4/3 camera and 25 mm lens. Image scale is established via an independently powered LIDAR/GPS data-logging system recording altitude and GPS location at 1 Hz. Photogrammetric calibration of the camera and lens allowed distortion parameters to be rigorously accounted for during image analysis, via a custom-programmed Graphical User Interface (GUI) running in MATLAB. The datalogger, camera calibration methods and measurement software are adaptable to a wide range Edited by: Alastair Martin Mitri Baylis, of UAV platforms. Mean LIDAR accuracy, measured from 10 bridges 9–39 m above South Atlantic Environmental water, was 99.9%. We conducted 136 flights in New Zealand’s subantarctic Auckland Research Institute, Falkland Islands Islands to measure southern right whales. Mean lengths of 10 individual whales, each Reviewed by: photographed between 7 and 15 times, had CVs (SD/mean) ranging from 0.5 to 1.8% Daniel Paul Costa, University of California, Santa Cruz, (mean = 1.2%). Repeated measurements of a floating reference target showed a United States mean error of c.1%. Our system is relatively inexpensive, easily put together, produces David Peel, Commonwealth Scientific and accurate, repeatable measurements from single vertical images, and hence is applicable Industrial Research Organisation to a wide range of ecological questions in marine and terrestrial habitats. (CSIRO), Australia *Correspondence: Keywords: aerial, photogrammetry, UAV, LIDAR, whale Stephen M. Dawson steve.dawson@otago.ac.nz INTRODUCTION Specialty section: This article was submitted to Photogrammetry is widely used in ecological research, often to provide measurements of animals Marine Megafauna, that are impractical or dangerous to capture (e.g., elephants, Rüther, 1982; sharks, Klimley and a section of the journal Brown, 1983; fin whales, Ratnaswamy and Winn, 1993; coelacanths, Décamps et al., 2016). Frontiers in Marine Science Unmanned Aerial Vehicles (UAVs) are increasingly used in ecological research (see Linchant et al., Received: 28 July 2017 2015, for review). Their use as photogrammetric platforms to measure mobile animals, however, is Accepted: 31 October 2017 a recent and evolving development. Published: 15 November 2017 The central challenge for photogrammetry from UAVs is establishment of scale, which, Citation: for mobile animals, must be achieved instantaneously. The simplest approach is to include Dawson SM, Bowman MH, in the image an object of known size, typically a boat (e.g., Whitehead and Payne, 1981; Leunissen E and Sirguey P (2017) Christiansen et al., 2016), or projected laser dots a known distance apart (e.g., Durban and Inexpensive Aerial Photogrammetry for Parsons, 2006; Rowe and Dawson, 2008). Having a boat in the image involves the potential Studies of Whales and Large Marine for disturbance, and our trials with lasers indicated that the dots may not show up on Animals. Front. Mar. Sci. 4:366. doi: 10.3389/fmars.2017.00366 whales at the ranges required, and raised questions about eye-safety. Alternatively, provided Frontiers in Marine Science | www.frontiersin.org 1 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals that camera separation and orientation are known precisely, a software tool we developed to measure images, and quantify stereo camera systems do not require an independent the accuracy and precision of resulting photogrammetric measurement of range. The extra weight and awkwardness measurements of a floating reference target and living whales. of a stereo camera system, however, are impractical on most A building guide for the datalogger, the code to run it, and our UAVs. If the subject can be assumed to be lying in a horizontal measurement software are provided as Supplementary Material. plane, vertical photography using a single gimballed camera, Our intention is to enable others to apply easily the potential of ideally with a low-distortion lens, and a means of accurately UAV-based aerial photogrammetry to their ecological research. measuring height, is far more practical. There are at least three approaches to measuring height. The METHODS AND RESULTS first uses data logged from a high-precision air pressure sensor Platform Choice (e.g., Durban et al., 2015). Provided there are no substantial The competing needs of light weight, high resolution, dynamic atmospheric changes during the flight, such data can accurately range and low distortion led us to the Micro4/3 system, and record altitude. Carrier-phase GPS is also an option; several to the Olympus 25 mm f1.8 lens. Rather than build a bespoke models are light enough to be carried on a medium-sized UAV, we chose to adapt the DJI Inspire 1 Pro (I1P) quadcopter UAV and can support centimeter accuracy. This approach for single-frame photogrammetry. This UAV has a gimballed relies on having an accurate terrain model, and, if over water, camera (X5, 16 MP, Micro4/3), and supports the Olympus 25 mm correcting for tidal height. LIDAR (Light Detection and Ranging) f1.8 lens. Live video feedback is displayed while in flight via an technology involves illuminating a target with pulsed laser light Apple iOS or Android device running DJI “GO,” which allows and measuring reflected pulses to provide information about adjustment of camera and gimbal settings. We use this particular target range. LIDAR was first applied in meteorology (e.g., UAV platform as an example only; the LIDAR datalogger, camera Goyer and Watson, 1963), but has found a very wide range of calibration process and measurement approach presented below applications since, including recent use in obstacle avoidance are applicable to a wide range of multirotor and fixed wing UAV and navigation in autonomous vehicles (e.g., Schwarz, 2010). platforms. Now that lightweight LIDAR units are available, they provide a simpler, practical solution for instantaneous measurement of altitude in UAVs. The system we describe here uses this LIDAR/GPS Datalogger The I1P has no ports giving access to its GPS data or power. technology. Our purpose is to provide an example by adapting an Therefore, we built an altitude measuring system that was completely independent (Figure 1). This comprised a Lightware off-the shelf “prosumer” quadcopter for specialized aerial photogrammetry, via inexpensive modifications. We provide SF11-C LIDAR, Pololu MinIMU-9 Inertial Measurement Unit, results of testing the accuracy and precision of the LIDAR unit, GlobalSat EM506 GPS module and 3.3 v 8 Mhz ProMicro describe the calibration of the UAV camera and lens, describe Arduino-compatible microcontroller (and MicroSD flash card FIGURE 1 | Block diagram of the datalogger. Text on the connecting lines shows communication protocols. Frontiers in Marine Science | www.frontiersin.org 2 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals FIGURE 2 | Components of the UAV photogrammetric system. breakout). The GPS module is used to timestamp the altitude While mean accuracy across replicated tests is very good, data so they can be matched to image timestamps. When the raw data showed occasional outliers, presumably due to the hovering, multirotor UAVs must tilt into the wind to maintain changing shape of the water surface. To investigate how these position. If the LIDAR is rigidly mounted to the UAV, its might be reduced via applying a median smoothing filter (which measurements of altitude will be biased high (by the cosine of the SF-11 can do internally), we conducted a further test from the the tilt angle). For this reason the datalogger system incorporates Clydevale bridge (9.77 m to water surface). We set the LIDAR to an Inertial Measurement Unit (IMU) which measures UAV tilt. output a moving median of 16 measurements (made at 16 Hz). We wrote code to integrate LIDAR measurements and IMU Both unfiltered and filtered data showed high accuracy (mean data into the output data sentences from the GPS module (GGL error +1.7 cm), but large differences in precision. Unfiltered data and RMC-NMEA protocols). Data are logged (at 1 Hz) by the showed maximum errors of +1.74 m and −1.42 m (n = 2508, SD microcontroller to a Micro SD card as .csv files. Power is supplied = 0.118 m) while the filtered data showed maximum errors of by a separate 7.4 v 500 mAH lithium-polymer battery, with its +0.06 m and −0.03 m (n = 2508, SD = 0.014 m). output regulated to 5 v using a 3 A UBEC. Combined weight of this system is <100 g. Camera Calibration System components are non-invasively mounted to the I1P To achieve rigorous scaling of distances to object size, via double-sided foam tape and gaffer tape. We built carbon- the camera’s interior orientation parameters (IOPs) and lens fiber brackets to support the GPS module and power system. distortion parameters must be known accurately. To avoid slight The latter is mounted at the aircraft nose to offset the weight of changes in focal length due to focussing, the I1P’s camera was components added behind the center of balance (see Figure 2). set to infinity focus. We calibrated the camera and lens by taking Detailed building instructions and the code to run the datalogger multiple aerial images of a field containing 88 40 cm stakes placed are available in Supplementary Material. 2 m apart in a 14 × 20 m grid pattern. A high-contrast target was fixed to the top of each stake. To support scaling, we also placed LIDAR Accuracy within the calibration field three precisely measured windsurfer The specifications of the SF11 LIDAR indicate ±10 cm accuracy masts each with a high-contrast target at each end. The I1P for ranges up to 40 m over moving water. To gain a more detailed was flown around the calibration field, stopping to take oblique evaluation, we mounted the LIDAR/GPS datalogging system on images every 45 degrees in a converging pattern, first at 20 m a wooden board, and clamped it to bridges ranging in height indicated altitude, then at an indicated 30 m and 40 m. This set above water from 9 to 39 m. Raw LIDAR measurements were was supplemented with nine vertical shots as the I1P was lowered recorded at 1 Hz for at least 60 s (n = 60–224). We were assured from 40 to 20 m above the center of the grid (Figure 3). The by the manufacturer that subsequent LIDAR measurements are resulting images were loaded into Australis photogrammetric independent. The Kawarau bridge was measured at two different software (www.Photometrix.com.au) in which stake-top targets heights above water (Table 1). were measured and scale defined. A free-net bundle adjustment Frontiers in Marine Science | www.frontiersin.org 3 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals TABLE 1 | Accuracy of raw SF-11 LIDAR ranges over water. Bridge True height (cm) Mean error (cm) % Error CV (SD/mean) (%) Balclutha 937 −1.4 −0.15 0.76 Clydevale 962.5 −3.0 −0.31 0.58 Horseshoe bend 1037 −0.1 −0.01 0.57 Cromwell 1107.5 −0.7 −0.06 1.75 Bannockburn 1128 −5.4 −0.48 2.43 Clyde 1278 −2.0 −0.16 0.97 Kawarau mining center 1604 −1.0 −0.06 0.72 Roxburgh 2002 −3.1 −0.15 0.64 Arthur’s point 2999 −4.4 −0.15 0.53 Kawarau B 3854 −4.0 −0.10 0.39 Kawarau A 3879 3.5 0.09 0.21 Averages −1.96 −0.14 0.87 True distance to water was measured using a weighted fiberglass tape measure. Outliers were not excluded. Vibration from passing traffic was discernable during tests from the Bannockburn and Cromwell bridges and may have contributed to their larger CVs. TABLE 2 | Calibration characteristics of our DJI X5 camera and Olympus 25 mm f1.8 lens according to the 10-parameter model employed in Australis (Fraser, 1997). Camera/lens parameter Abbrev Value Lens focal length fc 24.851372 mm Principal point on autocolimation PPA 0.203089 mm; −0.087931 mm Radial distortions k1 −9.1303 e-005 k2 8.4284 e-007 k3 −3.7862 e-009 Decentring distortions p1 −3.1598 e-005 p2 2.0922 e-005 Pixel scaling horiz/vertical b1 7.0190 e-004 Axial skew b2 −1.4177 e-004 image, applies the appropriate LIDAR measurement and camera calibration data, and prompts the user to indicate the total length FIGURE 3 | Image network used for camera calibration displayed in Australis of the animal (whale in our case) by creating a series of points subsequently to bundle block adjustment. (mouse clicks) from the tip of the jaw, along the spine, to the fluke notch. To allow for some imprecision, the software fits a smooth curve to these points. The software then offers options to measure other features (see Figure 4). The user is required to input data (Mikhail et al., 2001) was performed to establish the calibration on image quality, and can add notes. These data are written to characteristics of the camera and lens combination (Table 2). an Excel-formatted results file, and the software prompts for the The calibration yielded a hyperfocal length 0.6% shorter than next whale. The software also allows measurement of any other the specified focal length. The lens exhibited moderate barrel image of interest, via manually inputting a LIDAR measurement. distortion only, with maximum radial distortion of −56μm and The MATLAB code is available in the Supplementary Material, so marginal decentering distortion of 4μm at 11 mm radius (corner that others can adapt it for their specific needs. of the Micro4/3 sensor). If uncorrected, the combination of uncertainties on focal length and lens distortion would introduce Accuracy and Repeatability of a 1% error in the size estimate. Photogrammetric Measurements Graphical User Interface To measure accuracy of photogrammetric measurements, we We programmed a Graphical User Interface (GUI) in MATLAB took 65 photographs of a 2.716 m floating reference target, from to incorporate the measured camera calibration parameters, 16 to 30 m altitude, the usual range within which we photograph LIDAR ranges and tilt data into the image measurement process. whales, at the end of whale-measuring flights. On each occasion Using an Excel file listing best images, the GUI opens an the target was photographed in the center of the field of view, Frontiers in Marine Science | www.frontiersin.org 4 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals FIGURE 4 | Screenshot of the GUI measurement screen. The purple line is a curve fitted to user-defined points along the spine to measure total length. Yellow lines indicate where the user should click on the whale’s outline to measure body width at 10% intervals of total length. Green lines represent measurements from eye to eye, rostrum to eye rostrum to blowholes and fluke width. TABLE 3 | Measurements of total length (TL) of photographically identified right australis) for which we had at least seven different measurement- whales. grade images each (Table 3). Images had to be sharp, with the tip of the rostrum and fluke notch clearly visible, and the tail flukes Whale ID n Mean TL (cm) SD (cm) CV (SD/mean) (%) apparently flat (i.e., not drooped). AI16D026 7 1,041 6 0.5 The mean CV of total length measurements was 1.2%, AI16D014 15 1,338 9 0.6 surprisingly precise considering that whales are inherently AI16D025 9 1,285 9 0.7 flexible, there is inevitably some imprecision in tracing the spine, and that rippling of the water surface and refraction AI16D033 15 1,437 18 1.2 AI16D029 8 1,358 18 1.3 can compromise pointing accuracy. This CV implies a 95% confidence interval from a single image of a 14 m whale of AI16D017 9 1,325 18 1.4 13.66–14.33 m. There was no evidence of a relationship between AI16D050 12 1,401 20 1.4 CV and whale size. AI16D013 10 1,365 21 1.5 AI16D065 9 1,316 22 1.7 AI16D012 9 1,285 23 1.8 Deployment Good weather made it possible to fly the I1P on 12 days over ◦ ′ ◦ ′ 22 days in Port Ross (50 32 S, 166 14 E), a harbor at the and in each of the corners. Measurements showed positive bias Northern end of the Auckland Islands (26 July−17 August 2016). when the target was photographed in the bottom left corner and Seven flights were made from a 6.6 m rigid-hulled inflatable, slight negative bias when photographed in the top right corner landing on a 1.6 × 1.6 m platform. This proved challenging in (ANOVA, p = 0.015), suggesting that the UAV’s gimbal may not practice, so the remaining 129 flights were flown from the aft have been exactly vertical. Overall, mean error was 1% (2.7 cm, deck of the Otago University research vessel Polaris II. More SD = 4.9 cm). We made no attempt to correct for this small error recently, we have added “handles” to our UAVs (Figure 2), in measurements of whales. and shifted to hand-catching as our standard practice; this Repeatability of measurements was assessed from 10 makes flying from small boats much more practical. Flights photographically identified southern right whales (Eubaleana were generally 10–12 min in duration. To allow a generous Frontiers in Marine Science | www.frontiersin.org 5 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals safety margin, our protocol was to land with at least 40% sells as a complete system for $US20,000. In the field, the I1P of battery capacity remaining. Flights were made in winds of appeared to fly at least as well as the APH-22 hexacopter, and up to 20 knots, generally over smooth water in the lee of a was much easier to use (R.W. Wallace, APH-22 pilot, pers. shore. All appropriate permissions, including a research permit comm). from New Zealand’s Department of Conservation, were in place. Durban et al. (2015), whose APH-22 hexacopter estimates We saw no evidence of any reaction by the whales to the altitude from logged measurements of air pressure, report UAV. similar mean accuracy in measurements of fixed targets (mean Before safe flight, most UAVs must calibrate their Inertial errors; theirs = 0.7%, ours = 1.0%), but much worse precision Measurement Unit, and compass. Despite launching from (their CV = 4.5%; our CV = 1.8%). Their approach, and moving vessels, the I1P calibrated its IMU successfully on all but that of Christiansen et al. (2016), makes no allowance for one occasion, when the vessel was rolling. We occasionally had any lens/camera distortions. Six individual blue whales they difficulty calibrating the I1P’s compass. This may have been to do measured 4–7 times showed ranges of measurements within 5% with a local magnetic anomaly (strong enough to be marked on of the mean for each individual (Durban et al., 2016). This is the nautical chart of Port Ross), or to the substantial steel gantry a smaller sample (of whales and repeat measurements) than on the aft deck of Polaris II. Relatively close proximity (<2 m) we present above (Table 3), but suggests similar precision in to this was unavoidable. On the few occasions when reattempting actual use. This may be because precision is driven more by the calibration or moving the UAV a meter or so was not effective, the flexibility of whales than by errors in the photogrammetric we successfully calibrated the I1P’s compass on the wheelhouse process. Dawson et al. (1995) reached a very similar conclusion roof, away from magnetic materials. The Polaris II is a wooden comparing different boat-based methods of measuring sperm vessel. whales. We did notice an unexpected feature of the I1P’s image The largest contributor to photogrammetric error in our timestamps. These are synchronized from the tablet running system is LIDAR inaccuracy. Ranges measured from stable road “GO,” the camera control software. In remote areas, away from bridges showed occasional outliers. With the whale data, we access to a Network Time Protocol server, the tablet clock minimized outliers by applying a 5 s median filter to the LIDAR is likely to drift. We measured and corrected for this using measurements, centered on the time when the photograph was GPS timestamps of LIDAR data gathered during videoed take- taken. This is a reasonable because while photographing a whale, offs or landings. Time drift on our iPad Air was 9 s over the we seldom change throttle settings (therefore altitude). The 17 days on which we flew. In future remote trips we will consistency of measurements of known whales shows this was daily photograph the time screen of a handheld GPS with the effective. The LIDAR can be programmed to apply a median iPad. filter to raw measurements; enabling this feature increased the We also took along an APH-22 hexacopter equipped with an precision of altitude measurements by more than an order of Olympus E-PM2 and 25 mm f1.8 lens (Durban et al., 2015) and magnitude. two Swellpro Splashdrones equipped with Canon D30 waterproof This system was designed for measuring whales, but could point-and-shoot cameras programmed to take images every 2 s be applied to measuring any large animal swimming at the (Christiansen et al., 2016). The APH-22 flew well and collected surface (e.g., sharks, dolphins) or hauled out on a beach (e.g., data for 2 days, but crashed into the water when the velcro strap pinnipeds), or to measuring the spacing between individuals. It holding its battery in place failed. This incident was in contrast could also be used to measure habitat areas at small scales (e.g., to the usual reliability of this aircraft (e.g., Durban et al., 2015). coral patch reefs). The calculations assume that the object to be One of the Splashdrones flew briefly, and crashed into the water. measured is essentially on a flat plane whose range from the Thereafter neither could calibrate its flight controller, and hence camera is measured by the lidar. Without knowing exactly where did not fly. The local magnetic anomaly may have been the cause the lidar was measuring to, the system would not be suitable for of this problem. measurements of terrestrial animals on uneven ground. ETHICS STATEMENT DISCUSSION The research was conducted under permit 50094-MAR from The purpose of this contribution is to describe modifications Department of Conservation New Zealand. we have made to an off-the-shelf UAV to make it suitable for low-altitude aerial photogrammetry and provide sufficient information for others to be able to replicate it. These AUTHOR CONTRIBUTIONS modifications could be applied to a wide variety of mid- size UAVs. Excluding our development time, the cost in New SD developed the central ideas, conducted all fieldwork, analyzed Zealand of our I1P-based system, including 25 mm lens, iPad the data and led the writing process. MB designed and Air, 6 TB48 batteries and charger was approximately US$5,300. programmed the datalogger. EL developed the measurement The parts to build our GPS/LIDAR datalogging system total software. PS conducted the photogrammetric calibration. All approximately $US340. If you wish to use the GUI described authors contributed critically to the drafts and gave final approval above, you may need to buy MATLAB. The APH-22 hexacopter for publication. Frontiers in Marine Science | www.frontiersin.org 6 November 2017 | Volume 4 | Article 366 Dawson et al. Inexpensive Aerial Photogrammetry for Mobile Animals ACKNOWLEDGMENTS ltd) facilitated a discounted Inspire 1 Pro. Alastair Neaves, Mike Denham and Ray McLennan advised on measuring This research was principally funded by the New Zealand LIDAR accuracy. We thank Chris Fitzpatrick for administrative Antarctic Research institute, with additional support from assistance. Otago University and NZ Whale and Dolphin Trust. SD is especially grateful to Colin Gans for UAV advice. Mike SUPPLEMENTARY MATERIAL Paulin helped with our prototype LIDAR-datalogger. We thank The Supplementary Material for this article can be found online the crew and expedition members on board Polaris II for the 2016 Winter season. Richard Greasley (Lacklands NZ at: https://github.com/EvaLeunissen/Whalength REFERENCES Linchant, J., Lisein, J., Semeki, J., Lejeune, P., and Vermeulen, C. (2015). Are unmanned aircraft systems (UASs) the future of wildlife monitoring? Christiansen, F., Dujon, A. M., Sprogis, K. R., Arnould, J. P. Y., and Bejder, A review of accomplishments and challenges. Mammal Rev. 45, 239–252. L. (2016). Non-invasive unmanned aerial vehicle provides estimates of the doi: 10.1111/mam.12046 energetic cost of reproduction in humpback whales. Ecosphere 7:e01468. Mikhail, E. M., Bethel, J. S., and McGlone, J. C. (2001). Introduction to Modern doi: 10.1002/ecs2.1468 Photogrammetry. New York, NY: John Wiley & Sons. Dawson, S., Chessum, C. J., Hunt, P., and Slooten, E. (1995). An inexpensive, Ratnaswamy, M. J., and Winn, H. E. (1993). Photogrammetric estimates of stereo-photographic technique to measure sperm whales from small boats. Rep. allometry and calf production in fin whales, Balaenoptera physalus. J. Mammal. Int. Whal. Commn. 45, 431–436. 74, 323–330. doi: 10.2307/1382387 Décamps, T., Herrel, A., Ballesta, L., Holon, F., Rauby, T., Gentil, Y., et al. Rowe, L. E., and Dawson, S. M. (2008). Laser photogrammetry to determine dorsal (2016). The third dimension: a novel set-up for filming coelacanths in their fin size in a population of bottlenose dolphins from Doubtful Sound, New natural environment. Methods Ecol. Evol. 8, 322–328. doi: 10.1111/2041-210X. Zealand. Aust. J. Zool. 56, 239–248. doi: 10.1071/ZO08051 12671 Rüther, H. (1982). Wildlife stereo photogrammetry at close range. Int. Arch. Durban, J. W., Fearnbach, H., Barrett-Lennard, L. G., Perryman, W. L., Photogramm. 24, 422–432. and Leroi, D. J. (2015). Photogrammetry of killer whales using a small Schwarz, B. (2010). LIDAR: mapping the world in 3D. Nat. Photon. 4, 429–430. hexacopter launched at sea 1. J. Unmanned Veh. Syst. 3, 131–135. doi: 10.1038/nphoton.2010.148 doi: 10.1139/juvs-2015-0020 Whitehead, H., and Payne, R. (1981). New Techniques for Measuring Whales from Durban, J. W., Moore, M. J., Chiang, G., Hickmott, L. S., Bocconcelli, A., Howes, the Air. US Marine Mammal Commission Report MMC-76/22. Washington, G., et al. (2016). Photogrammetry of blue whales with an unmanned hexacopter. DC, 36. Mar. Mammal Sci. 32, 1510–1515. doi: 10.1111/mms.12328 Durban, J. W., and Parsons, K. M. (2006). Laser-metrics of free ranging Conflict of Interest Statement: The authors declare that the research was killer whales. Mar. Mammal Sci. 22, 735–743. doi: 10.1111/j.1748-7692.2006. conducted in the absence of any commercial or financial relationships that could 00068.x be construed as a potential conflict of interest. Fraser, C. S. (1997). Digital camera self-calibration. ISPRS J. Photogramm. Remote Sens. 52, 149–159. doi: 10.1016/S0924-2716(97)00005-1 Copyright © 2017 Dawson, Bowman, Leunissen and Sirguey. This is an open-access Goyer, G. G., and Watson, R. (1963). The laser and its application to meteorology. article distributed under the terms of the Creative Commons Attribution License (CC Bull. Am. Meterol. Soc. 44, 564–575. BY). The use, distribution or reproduction in other forums is permitted, provided the Klimley, A. P., and Brown, S. T. (1983). Stereophotography for the field biologist: original author(s) or licensor are credited and that the original publication in this measurement of lengths and three-dimensional positions of free-swimming journal is cited, in accordance with accepted academic practice. No use, distribution sharks. Mar. Biol. 74, 175–185. doi: 10.1007/BF00413921 or reproduction is permitted which does not comply with these terms. Frontiers in Marine Science | www.frontiersin.org 7 November 2017 | Volume 4 | Article 366

Journal

Frontiers in Marine ScienceUnpaywall

Published: Nov 15, 2017

There are no references for this article.