Development of an Unmanned Aerial Vehicle (UAV) for hyper resolution... mapping based on visible, multispectral, and thermal imagery

Document technical information

Format pdf
Size 2.5 MB
First found May 22, 2018

Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

Persons

J. R. Miller
J. R. Miller

wikipedia, lookup

Organizations

Places

Transcript

Development of an Unmanned Aerial Vehicle (UAV) for hyper resolution vineyard
mapping based on visible, multispectral, and thermal imagery
D. Turner a,*, A. Lucieer a, C. Watsona
a
School of Geography and Environmental Studies, University of Tasmania, Hobart, 7001, AUSTRALIA
(Darren.Turner, Arko.Lucieer, Christopher.Watson)@utas.edu.au
Abstract – Unmanned Aerial Vehicles (UAVs) are an
exciting new remote sensing tool capable of acquiring high
resolution spatial data. This study has developed a UAV
capable of collecting hyper resolution visible, multispectral
and thermal imagery for application to Precision Viticulture
(PV). Traditional modes of data collection are not well
suited to the detection of subtle but important changes in
vineyard structure given low temporal and spatial
resolutions. Mapping with UAVs has the potential to
provide imagery at an unprecedented spatial and temporal
resolution. We present a technical description of our UAV
and its payload options including visible imagery which is
processed using feature matching and photogrammetric
techniques to create Digital Surface Models (DSMs) of the
vineyards. A thermal infrared camera is used to map soil
moisture enabling assessment of irrigation efficiency, and a
six-band multispectral camera enables the calculation of
vegetation indices that relate to vineyard vigour and health.
Keywords: UAV, Precision Viticulture,
Vegetation Indices, Thermal Infrared
1.
bio-geophysical properties of the vines such as Leaf Area Index
(LAI).
Proffitt and Pearse, (2004) used ground truth data to
demonstrate that PCD provides a good indication of vine vigour
that correlates well to yield and wine quality. Berni et al.(2009)
discovered that the Photochemical Reflective Index is a much
better indication of plant water stress than NDVI when looking
at orchard crops such peaches and olives.
However, vineyards are not a homogeneous crop and data from
traditional aerial photography or high resolution satellite sensors
are not able to accurately differentiate between individual vines
and the between row vegetation. This dictates that at the edges
of the vines, pixels will exhibit the reflectance properties of
both the vine itself and the adjacent vegetation. Hyper
resolution UAV imagery of vineyards does not have this
limitation, hence there is an exciting potential for development
of a UAV system to map vineyards and thus assist with PV.
Multispectral,
INTRODUCTION
Historically, Unmanned Aerial Vehicles (UAVs) have primarily
been used for military applications. More recently, the use of
UAVs in the civilian domain as remote sensing tools presents
new and exciting opportunities. Improvements in the
availability of accurate and miniature Global Positioning
System (GPS) units and Inertial Measurement Units (IMUs),
along with the availability of quality off the shelf consumer
grade digital cameras has resulted in an increased use of civilian
UAVs (Nebikera et al., 2008).
The highest spatial resolution data available from conventional
platforms such as satellites and manned aircraft is typically in
the range of 20-50 cm/pixel. UAVs are capable of flying much
lower and hence can collect imagery at a much higher resolution
(Scaioni et al., 2009; Hunt et al., 2010), often at a sub-decimetre
resolution, even as detailed as 1 cm/pixel.
This paper describes our UAV platform and the sensors that are
used to collect data. We demonstrate the processing algorithms
that are used to process and ortho-rectify data from each of our
sensor types. Some vegetation indices will then be examined to
qualitatively assess their ability to differentiate differences
within vineyard imagery.
2.
METHODS
Our UAV is based on the Oktokopter platform
(Mikrokopter, 2011), a multi-rotor electric powered system
purpose designed for aerial photography (Figure 1). The
Oktokopter has been fitted with a stabilised camera mount to
which we mount our sensor systems. The Oktokopter has a
payload limit of approximately one kilogram, hence we are
limited to flying each of our sensors individually.
The temporal resolution of conventional systems is limited by
the availability of aircraft platforms and orbit coverage patterns
of satellites. For the purpose of monitoring highly dynamic
vegetation such as that within vineyards, satellite sensors are
very much limited due to unfavourable re-visit times
(Berni et al., 2009).
Precision Viticulture (PV) is defined as monitoring and
managing spatial variations in physical, chemical, biological
variables related to productivity of vineyards (Hall et al., 2002).
Extensive research has been undertaken to determine if
traditional sources of remote sensing data (satellite and aerial
photography imagery) can be used to assist with PV,
Hall (2008) provides one example. Much of this research has
been based around calculating vegetation indices such as Plant
Cell Density (PCD) and Normalised Difference Vegetation
Index (NDVI), and subsequently relating these to
Figure 1. Oktokopter fitted with multispectral camera.
A Canon 550D digital SLR camera is used to capture visible
imagery which is then processed with feature matching and
photogrammetric software to create Digital Surface Models
(DSMs). The same camera mount has been adapted to carry a
Tetracam mini-MCA (Multi Camera Array) (Tetracam, 2011)
multispectral camera that operates in six bands set by fitting
specific filters. This imagery is used to examine vegetation
reflectance in critical wavelengths allowing the calculation of
vegetation indices. Finally, a FLIR (FLIR, 2011) Thermal
Infrared (TIR) camera can be attached to the Oktokopter to
measure surface temperature.
The Oktokopter has an onboard navigation system based on a
GPS receiver and an IMU. A defined set of waypoints can be
pre-programmed to form a set flight path. Flight planning
software has been developed that calculates the spacing and
layout of waypoints to acquire data over a region of interest at a
nominated image scale (Figure 2). Imagery is typically acquired
at the maximum rate allowed on the device, thereby providing
redundancy to account for elimination of imagery with
excessive tilt, motion blur or bad exposure.
matched feature is listed along with its RGB colour from the
original imagery.
These bundler point clouds are sparse, however, denser point
clouds but can be generated by use of PMVS2 software
(Furukawa and Ponce, 2009) to produce point clouds containing
many millions of points. (Figures 3 and 4)
Figure 3. Vineyard point cloud, before vine canopy growth
Figure 2. Flight planning software. Flight grid is approximately
200 x 100 m which represents a single flight sequence
(5 mins flight duration)
3.
TEST CASES
We have tested our UAV system at the Frogmore Creek
vineyard near Richmond in southern Tasmania. This is a large
vineyard containing many blocks of vines of different varieties.
We have identified one block of particular interest for use in
this study – it consists of highly variable topography that affects
the surface moisture and thus is one of the influences the vigour
of the vines.
A significant challenge associated with UAV imagery is the
large volume of data that is produced; to deal with this
automated processing systems are required. We have developed
a semi-automated system to ortho-rectify the imagery with or
without Ground Control Points (GCPs).
To ortho-rectify the imagery we initially manually select the
most appropriate photos from the dataset collected. The selected
images are processed with the Bundler software (Snavely, 2010)
which uses the SIFT algorithm (Lowe, 2005) to detect matching
features across the images.
Bundler then runs a bundle block adjustment to align the images
within its own coordinate system. An output file is generated
that lists the calculated position of each exposure station in the
bundler coordinate system along with the location of each
matched feature in both the bundler and original image based
coordinate system. A point cloud is also produced in which the
x, y, z position (in the bundler coordinate system) of each
Figure 4. Vineyard point cloud, vines with full canopy
If GCPs are present in the imagery we use a simple filtering
algorithm (based on thresholding to identify the bright orange
colour of the control points) to extract the location of the GCPs
from the point cloud and subsequently match them to the
locations measured in situ with survey-grade, carrier phase
differential GPS equipment. This provides a list of x, y, z points
in the bundler coordinate system with a corresponding northing,
easting and elevation coordinate in a real world coordinate
system, in our case the Map Grid of Australia (MGA). The
point cloud is transformed into MGA following the solution for
Helmert transformation parameters (three translations, three
rotations and one scale parameter) using a least squares
algorithm.
When no GCPs are present in the image the camera locations in
the Bundler output file are matched to the onboard GPS logs
from the Oktokopter based on the time at which the image was
captured. This again gives us a list of coordinates matched to
their real world equivalents allowing Helmert transformation
parameters to be determined. This technique is presently limited
to the inaccuracy of the onboard GPS unit that is a navigation
grade single frequency GPS device.
Initial validation of the point cloud transformed with the aid of
GCPs yields an RMS error of approximately 10 cm. Without
GCPs, the RMS increases to ~2 m given the quality of the GPS
unit on board.
The calculated Helmert parameters can also be applied to the
Bundler output file to create control points for each image. That
is, for each matched feature in an image we have an x, y, z
coordinate which can be transformed into a northing, easting
and elevation coordinate. For a given matched feature, we also
have an image coordinate, this means there is now a list of
control points for each image (usually in the order of thousands)
for which we have real world 3D coordinates (Figure 5)
content. Figure 7 provides a TIR example that highlights a
temperature gradient up to the bottom right of the image where
temperatures are higher than in other areas. Inspection of the
study region shows that this region is less vigorous (both vines
and between row vegetation), hence more soil is exposed and
the ground is warmer.
30m
50m
Figure 5. Control points from bundler shown in red.
These control points can now be used with a rubber sheet
transformation to orthorectify each image. Once all images are
orthorectified it is a simple task to mosaic them together with a
geo-referenced based stitching software such as ENVI
(ITTVIS, 2011). (Figure 6)
20°C
Temperature
40°C
Figure 7. Example TIR image of vineyard
Data from the multispectral camera can be viewed as a false
colour composite. We use this to generate a set of “quicklooks”
allowing us to find regions of specific interest. Figure 8 is of the
same area as the TIR image and demonstrates that the less
vigorous area is also well defined in the multispectral imagery.
30m
Figure 8. False colour composite of vineyard area where
Red = 800 nm, Green=670 nm and Blue=550 nm
Figure 6. Example ortho-mosaic of visible vineyard imagery
made up from 48 images. Area is approximately 200x150m
Further data processing algorithms have been developed to
convert the raw data collected by the multispectral and TIR
camera into usable imagery. The TIR images indicate the soil
and surface temperature which is generally related to moisture
A commonly used vegetation index is NDVI which provides a
general measure of vegetation vigour and biomass. NDVI is a
ratio and has the advantage that an object under shadow will
reflect light reduced by the same amount across the entire
spectrum, thus it is not severely affected by shadows in the
image (Hall, 2002). NDVI is commonly used in PV
applications, for example, Lamb et al. (2001) found a
correlation between image derived NDVI values and subsequent
grape yields. Figure 9 displays the NDVI generated from the
multispectral image display in Figure 8, again the less vigorous
area is well defined.
Further studies will be undertaken to determine how these
datasets can best be used by vineyard managers to implement
decisions that improve operational efficiency, productivity and
sustainability.
30m
Figure 9. NDVI for same section of the vineyard
Another commonly used vegetation index is the Photochemical
Reflectance Index (PRI) that is an indicator of chlorophyll
fluorescence which changes under water stress conditions
(Suárez et al., 2008). Berni et al (2009) demonstrated that PRI
generated from UAV multispectral imagery could be used for
water stress detection. Figure 10 shows how PRI can be used to
differentiate between the vines and the background surface. It is
possible to see that the vines in the upper left are vigorous and
leafy (i.e. brighter in the PRI image) whilst in the bottom right
there are far fewer leaves and areas where you can see the trellis
wires in the absence of any significant vines.
15m
Figure 10. PRI for zoomed in section of the vineyard
4. CONCLUSIONS
These initial results presented in this study further highlight the
potential for multi-sensor UAV systems in the application of
PV. The versatility of the UAV system is further enhanced by
the fact that these data sets can be collected “on-demand”,
providing unprecedented temporal resolution that spans the
critical times in the crop growing season.
The imagery produced from UAV collected data in this study
has a spatial resolution up to 1 cm/pixel. The accuracy of the
ortho-mosaics produced depends on the use of GCPs, however,
a dual frequency survey grade onboard GPS has the potential to
eliminate the need for GCPs and thus greatly improve the
efficiency of the UAV field to finish system.
5. REFERENCES
Berni, J. A. J., P. J. Zarco-Tejada, L. Suarez and E. Fereres
(2009). "Thermal and Narrowband Multispectral Remote
Sensing for Vegetation Monitoring from an Unmanned
Aerial Vehicle." IEEE Transactions on Geoscience and
Remote Sensing 47(3): 722-738.
FLIR (2011) “Thermal Imaging, Infrared Camera Systems”
http://www.flir.com
Furukawa, Y. and J. Ponce (2009). "Accurate, Dense, and
Robust Multi-View Stereopsis." IEEE Trans. on Pattern
Analysis and Machine Intelligence.
Hall, A., D. Lamb, B. Holzapfel and J. Louis (2002). "Optical
remote sensing applications in viticulture - a review."
Australian Journal of Grape and Wine Research 8: 36–47.
Hall, a., J. P. Louis, et al. (2008). "Low-resolution remotely
sensed images of winegrape vineyards map spatial
variability in planimetric canopy area instead of leaf area
index." Australian Journal of Grape and Wine Research 14:
9-17.
Hunt, J. E. R., W. D. Hively, S. Fujikawa, D. Linden, C. S.
Daughtry and G. McCarty (2010). "Acquisition of NIRGreen-Blue Digital Photographs from Unmanned Aircraft
for Crop Monitoring." Remote Sensing 2(1): 290-305.
ITTVIS (2011) “ENVI Software - Image Processing and
Analysis Solutions” http://www.ittvis.com/envi
Lamb, D., A. Hall and J. Louis (2001). "Airborne remote
sensing of vines for canopy variability and productivity."
Australian Grapegrower and Winemaker: 89–94.
Lowe, D. (2005). SIFT Keypoint Detector.
Mikrokopter
(2011)
“Mikrokopter
WIKI”
http://www.mikrokopter.com
Nebikera, S., A. Annena, M. Scherrerb and D. Oeschc (2008). A
Light-weight Multispectral Sensor for Micro UAV Opportunities for very high resolution airborne remote
sensing. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, Beijing.
Proffitt, T. and B. Pearse (2004). Adding value to the wine
business precisely: using precision viticulture technology in
Margaret River. The Australian & New Zealand
Grapegrower & Winemaker. December: 5.
Rango, A., A. Laliberte, J. E. Herrick, C. Winters, K. Havstad,
C. Steele and D. Browning (2009). "Unmanned aerial
vehicle-based remote sensing for rangeland assessment,
monitoring, and management." Journal of Applied Remote
Sensing 3(033542): 15.
Scaioni, M., L. Barazzetti, R. Brumana, B. Cuca, F. Fassi and F.
Prandi (2009). RC-Heli and Structure & Motion techniques
for the 3-D reconstruction of a Milan Dome spire
Proceedings of the 3rd ISPRS International Workshop 3DARCH 2009: "3D Virtual Reconstruction and Visualization
of Complex Architectures" Trento, Italy.
Snavely, N. (2010). Bundler: Structure from Motion (SfM) for
Unordered Image Collections.
Suárez, L., P. J. Zarco-Tejada, G. Sepulcre-Cantó, O. PérezPriego, J. R. Miller, J. C. Jiménez-Muñoz and J. Sobrino
(2008). "Assessing canopy PRI for water stress detection
with diurnal airborne imagery." Remote Sensing of
Environment 112.
Tetracam (2011) “Tetracam Inc. Digital Camera and
Imaging Systems Design – Mini Multiple Camera Array”,
http://www.tetracam.com/mca_mini.html
×

Report this document