Second International workshop:
Image analysis for soil structure characterization
18-20 January 2016
Auckland, New Zealand
PROCEEDINGS
This project has received funding from the European Union’s Horizon 2020 research
and innovation programme under grant agreement N°645717.
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Contents
Programme ................................................................................................................................ 3
PROTINUS project and advances in image analysis for soil structure .............................. 4
Céline Duwig
Effect of soil structure on copper mobility in Andisols ....................................................... 5
Karin Müller
An integrated 3D computer vision solution for time-lapsed underwater imaging. ........... 6
Valentin Baron
An image processing pipeline for environmental sciences applications .......................... 7
Patrice Delmas
A multi-beam LIDAR system for Unmanned Aerial Vehicles ............................................... 8
Wannes van der Mark
Imagery in hydraulic engineering research ........................................................................... 9
Heide Friedrich
Using laser scanning to investigate sedimentation patterns in the field: a UK case
study ........................................................................................................................................ 10
Jane Groom
A mobile application for CT-scan image segmentation benchmarking: Demo and usercase study ............................................................................................................................... 11
Luke Chang
Soil management, soil porosity, gas diffusion and nitrogen transformations ................ 12
Steve Thomas
Analysis of soil functions based on pore scale imaging: Computation of unsaturated
properties of a sandy soil ...................................................................................................... 13
Laurent Oxarango
3D images for modelling behavior of natural porous media in geosciences .................. 14
Anne-Julie Tinet
Reconstruction of 3-D Scenes Captured by UAV using a Stereo Camera System ......... 15
Trevor Gee
2
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Programme
“Image analysis for soil structure characterisation”
Monday 18/01/2016 (Location: 303-412, Maths department, level 4, room 412)
11h00: Registration
11h30: Welcome speech (Celine Duwig-Patrice Delmas - HoD CompSci (Gill Dobbie)
12h15: Lunch break
13h15: Keynote: Celine Duwig (IRD Grenoble, France): “PROTINUS project presentation”
14h00: Karin Muller (Plant and Food Researhc, Hamilton, New Zealand): “Effect on soil
structure on copper mobility in Andisols”
14h45: Valentin Baron (UoA Computer Science, Auckland): “An integrated 3D computer
vision solution for time-lapsed underwater imaging”
15h30: Afternoon tea
16h00: Patrice Delmas (UoA Computer Science, Auckland): “IP4ES (Image Processing for
Environmental Sciences): The image processing pipeline”
16h45: Wannes van der Mark (UoA Computer Science, Auckland): “Multi-beam LIDAR
system for Unmanned Aerial Vehicles”
17h30: Closing
18h30-20h30: Welcome Cocktail (Sky Café-Sky Tower)
Tuesday 19/01/2016 (Location: 303s-561, CompSci department, level 5, room
561)
9h30: Heide Friedrich (UoA Civil & Environmental Engineering, Auckland): “‘Imagery in
hydraulic engineering research »
10:15: Morning Tea
10h45: Jane Groom (UoA Civil & Environmental Engineering, Auckland): “Using laser
scanning to investigate sedimentation patterns in the field: a UK case study »
11h30: Luke Chang (UoA Computer Science, Auckland): “A mobile application for Ct-scan
image segmentation benchmarking: Demo and user-case study”
12h30: Lunch
13h30: Steve Thomas (Plant and Food Research, Christchurch, New Zealand): “Soil
management, soil porosity, gas diffusion and nitrogen transformations”.
14h15 : Laurent Oxarango (LTHE, Grenoble University, France): “Understanding transfers in
porous media: Benefit and open questions around 3D Pore scale imaging”
15h00: Anne-Julie Tinet (UoL Georessource Nancy, France): ”Identifying today’s best
practices in 3D soil modelling at pore scale”
15h45: Trevor Gee (UoA Computer Science, Auckland): “Reconstruction of 3-D Scenes
Captured by UAV using a Stereo Camera System”
16h15: Closing Speech
16h30: Afternoon Tea
Wednesday 20/01/2016
PROTINUS internal meetings 9.30 am-5 pm
3
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
PROTINUS project and advances in image analysis for soil structure
Duwig C1*
1
IRD/Université Grenoble Alpes /CNRS/G-INP, LTHE UMR 5564, F-38000 Grenoble, France.
* Corresponding author: [email protected]
In the Framework of PROTINUS, our network supported by H2020 Marie Sklodowska-Curie
Research and Innovation staff exchange, we are pleased to welcome you to our second
international workshop hold in Auckland, New Zealand. The aim of this workshop is to get
together experts in the fields of image analysis adapted to soil science and image
processing.
After a short introduction on the different fields of soil science, and the importance to study
soils and its structure, I will pass in review the different progress we have made in the past
few years in the field of image analysis for soil science.
Our first work was done on thin layers of soil cores observed through microscopes, where we
could study numbers and characteristics of macropores (Prado et al., 2009) and compare
them to the transport of a water tracer in the same soil cores. To visualise the real macropore
flow pathways, we set up an experiment with a fluorescent tracer through a big soil cores,
which was afterwards cut and photographed, to be able to reconstruct these pathways in 3D
(Duwig et al., 2008). We coupled different techniques to infer soil porous structures, using
classical indirect techniques such as mercury intrusion porosimetry and direct visualisation of
the macropores network by Computer Tomography (CT) (Dal Ferro et al., 2012). CT was
also used to visualise directly the movement of water in a double porosity media (Peng et al.,
2015), or the impact of macrofauna on soil porosity (Prado et al., 2016). The reconstructed
porous network visualised by CT is being used to model water movement using a lagragian
method based on Smoothed Particle Hydrodynamics method (Dal Ferro et al, 2015).
References:
Dal Ferro N., Gastelum Strozzi A., Duwig C., Delmas P., Charrier P., Morari F., 2015.
Integrating X-ray computed microtomography and smoothed particle hydrodynamics (SPH)
to predict saturated water conductivity in a silty loam Cambisol. Geoderma, 255–256: 27-34.
Dal Ferro N., Delmas P., Duwig C., Simonetti G., Morari F., 2012. Coupling X-ray
microtomography and mercury intrusion porosimetry to quantify aggregate structures of a
Cambisol under different fertilisation treatments. Soil Tillage, 119 :13–21.
Duwig C., Delmas P., Müller K., Prado B., Morin H., Ren, K., 2008. Quantifying fluorescent
tracer distribution in allophanic soils to image solute transport. European Journal of Soil
Science 59, 94-102.
Peng Z., Duwig C., Delmas P., Gaudet J.P., Gastelum Strozzi A., Charrier P., Denis H.,
2015. Visualization and characterization of heterogeneous water flow in double-porosity
media by means of X-rays Computed Tomography. Transport in Porous Media, 110:543–
564.
Prado B., Gastelum Strozzi A., Huerta E, Duwig C., Zamora O., Delmas P., Casasola D.,
Márquez J., 2016. 2,4-D mobility in two clay soils: impact of macrofauna abundance and soil
functional groups. In revision in Geoderma.
Prado B., Duwig C., Márquez J., Delmas P., Morales P., James J., Etchevers J., 2009.
Image Processing-based study of soil porosity and its effect on water movement through
Andosol intact columns. Agricultural Water Management, 96 (10), 1377-1386.
4
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Effect of soil structure on copper mobility in Andisols
Müller K1*, Duwig C2, Spadini L3, Morel MC3,4, Gastelum Strozzi A5, Delmas P6, Clothier
B7
1*
Plant & Food Research, East Street, Hamilton 3214, New Zealand
IRD/Université Grenoble 1/CNRS, LTHE UMR 5564, Grenoble, France
3
Université Grenoble 1/CNRS/ IRD, LTHE UMR 5564, Grenoble, France
4
Conservatoire National des Arts et Métiers, 292 rue Saint Martin 75141 Paris cedex 03,
France
5
G-INP/Université Grenoble 1/ IRD, LTHE UMR 5564, Grenoble, France
6
The University of Auckland, Department of Computer Science, Auckland, New Zealand
7
Plant & Food Research, Private Box 11 600, Palmerston North 4442, New Zealand
* Corresponding author: [email protected]
2
Sustainable horticulture depends on the integrity of various soil functions, which in turn
directly depend on soil structure affecting aggregation, root growth, soil biota, and liquid and
gas permeability. A necessary step to develop eco-efficient horticultural systems is to identify
how soil structure can be preserved and manipulated by orchard management strategies.
We hypothesized that changes in the structure of soil pore networks resulting from the
feedback mechanisms between organic carbon management practices and soil biota can be
captured with 3D X-ray computed tomography (CT), and that such changes will lead to
differences in the regulating ecosystem services that soils are providing. Our particular
interest in this study was the filtering function of soils. We compared soils that have been
under organic and integrated kiwifruit production in the Bay of Plenty, New Zealand for 30
years. Copper was chosen as model for trace metallic elements. Generally, copper is widely
used to control fungi and bacterial diseases in orchards. Worldwide, this led to concerns
about the potential accumulation of copper in orchard soils because copper can be toxic to
soil organism depending on its bioavailable concentration and negatively affect the entire
plant-soil-food web. The specific motivation for our study was the current increased
application of copper in New Zealand kiwifruit orchards as response to the recent incursion of
Pseudomonas syringae pv. actinidiae that is threatening the kiwifruit industry of the country.
The 3D-macro pore networks and morphological parameters of undisturbed topsoil cores
from the vine-rows of both orchards were derived with X-ray CT combined with image
processing techniques. Then, a pulse of bromide and copper was leached through the same
cores. Leachate samples were analyzed for both solutes. After leaching of three pore
volumes, the soil cores were cut into slices and analyzed for total copper, organic carbon and
nitrogen contents, and soil pH. We discuss the correlation between the 2D-distribution of
total copper with parameters of the pore network derived from the CT images and other soil
parameters, as well as its impact on copper mobility. Preliminary results revealed that
macropores were on average larger, had a higher connectivity, but were significantly
(P<0.05) fewer in the organic than the integrated orchard topsoil. We hypothesized that this
unexpected result was due to the effect of herbicides on the shallow kiwifruit roots in the
integrated orchard. Generally, preferential water flow and transport of copper occurred in the
topsoil of the integrated orchard, while water flux was homogeneous and copper was
immobilized in the top 2 cm of the organic orchard soil. These results provide orchardists
information on the fate of copper in their soils.
5
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
An integrated 3D computer vision solution for time-lapsed underwater imaging.
Baron V1*, Gee T2, Delmas P3
1
Engineering student in ENSE3, Grenoble INP, Grenoble 38000, FRANCE
IVS Lab, The University of Auckland, Auckland 1010, New-Zealand
3
IVS Lab, The University of Auckland, Auckland 1010, New-Zealand
* Corresponding author
2
Interdisciplinarity spreads in science fields nowadays. It is the basis for innovation in
research and it allows the fast development of each of the fields. This project fits in this
context by allowing Marine science students to analyze results from stereo vision to be able
to describe better the life and health of an entire bay. Indeed, we focus here on little amounts
of sand created by worms in Leigh bay which describe very well the bay health. Nevertheless
to study them more accurately and follow their dynamic, computer computations have to be
performed and in particular stereo vision. The IVS Lab provides this study by giving students
the tools to take underwater pictures of the amounts thanks to a stereo pair of cameras,
sending them to a website owned by the computer science department, processing the
images to create 3D point clouds and then sending back the results to the students
concerned. As a starting point, the question of taking pictures underwater is interesting. We
use a stereo pair of GoPro Hero 3+ in its waterproof box to handle that. Then we tie it to a
smartphone thanks to a coax cable to be able to send wifi signal from the phone to the
cameras. Once this is installed, you can take pictures every two minutes by using our
Android application which leads us until the end of the tide, so six hours of images. You can
send these images on the website made for this project and finally process them to recover a
dynamic 3D point cloud in your screen. This point cloud is displayed thanks to a WebGL
window on each browser recent enough and along it you have a volume estimation of this
amount at each step of its dynamic. So we provide a full solution to help another field to
analyze their data but they have helped us to improve our algorithms for this particular
project too, this is the emulation of interdisciplinarity.
6
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
An image processing pipeline for environmental sciences applications
Delmas P1*, Chang L1, Gastelum Strozzi A2, Marquez Flores J2
1*
Department of Computer Science, The University of Auckland, Room 384, Level 3, Science
Centre Building 303S, 38 Princes Street, Auckland, New Zealand
2
CCADET, Circuito Exterior s/n, C.P. 04510, CD. Universitaria, Mexico, D.F., Mexico
* Corresponding author: [email protected]
New advances in medical imaging technologies offer reliable ways to visualize pore networks
in undisturbed soil cores while steady increases in processing power now allow to
seamlessly characterize pore networks structure in 3D. An image processing pipeline
encompassing imaging, image processing, and parameters estimation is presented. The
imaging section rolls back to past and current visualization techniques from 2D photography
of thin slices, fluorescent imaging to today’s standard CT-scan technologies. First, Data
normalisation solutions, such as histogram peak alignment or bleach correction using
histogram matching are presented. Next, pre-processing image filtering solutions to remove
noise and artifacts in CT-scan data are exemplified for Median and Gaussian filters. Image
segmentation options for unimodal and bi-modal images or stack of images are
demonstrated including Unimodal and Isodata thresholding of Andosol cores. Pore network
labelling and visualization solutions, including percolating pores and REV determination, are
presented using ImageJ and Matlab in-house code. Finally, a wealth of static and dynamic
2D and 3D pore network parameters and relevant formulas are summarized. The complete
pipeline is demonstrated using a mix of in-house and freely available ImageJ plugins.
Fig 1. The image processing pipeline. From left to right: Data before (top) and after histogram peak
normalisation, an image slice from an Andosol soil core, binarised image after unimodal thresholding,
3D pore network visualisation from 100 segmented images.
7
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
A multi-beam LIDAR system for Unmanned Aerial Vehicles
van der Mark W1* , Delmas P1
1*
Department of Computer Science, The University of Auckland, Auckland 1010, New
Zealand (NZ)
* Corresponding author
Unmanned Aerial Vehicles (UAV) have become a popular and cost effective method for
remote sensing from an aerial viewpoint. In this talk we will present a multi beam LIDAR
system that was integrated into an octa-copter UAV. LIDAR is a sensor technology that uses
light to measure distance. By measuring the returned light over time different outdoor
geometries can be identified such as the difference between the top of a forest canopy and
the soil surface. Areas can be scanned in 3D; however, conventional ground based LIDAR
systems are limited by the time required to cover large areas. We use the novel Velodyne
LIDAR sensor that utilizes multiple light emitters and receivers to enable real-time 3D
scanning. Sensor data is stored on a small embedded computer on board the UAV. To
accuracy annotate the LIDAR data with position data the system has been extended with
Inertial Navigation Sensors (INS) such as accelerometers, gyroscope, barometer and a
compass. The GPS system uses a low cost solution that is capable of real-time kinetic (RTK)
accuracies. We will also outline our future work such as using the LIDAR data itself for
localization during flight in order to build high accuracy 3D maps of outdoor environments.
8
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Imagery in hydraulic engineering research
Friedrich H1
1
University of Auckland, New Zealand
We experience the world around us in four dimensions: 3D scenes changing over time. Our
vision, together with hearing, touching, tasting and smelling, is important in how we relate
with our environment. In hydraulic engineering we encounter numerous challenges
conveying the abstract or not easily visible to something measurable, not only visualising, but
also quantifying.
When studying natural water-worked environments, such as rivers, often two media - one
solid and in the case of a riverbed composed of individual grains (sediment), the other liquid
(water) - transform concurrently; this fluid-boundary process is currently not fully described
mathematically. Clever experimental setups are needed to gain greater insights into the
physics of fundamental hydraulic processes. In this talk I presented the work my research
group has been done over the last years, using our own innovative imaging technologies in
the hydraulic laboratory environment to spatially and temporally decompose hydraulic
processes to study phenomena that take place in water-worked environments (Figure 1).
Figure 1. Fine-sediment bedform pattern (left); near-boundary flow measurement (middle);
sediment-laden flow (right).
9
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Using laser scanning to investigate sedimentation patterns in the field: a UK
case study
Groom J1
1
Department of Civil and Environmental Engineering, University of Auckland, New Zealand.
Within a fluvial environment, changes to the system can disrupt feedback mechanisms. After
several failed attempts at river management, the River Wharfe, UK, is experiencing high
levels of sedimentation, resulting in increased flood risk and bank erosion. A study into a
6.5km reach of the river analysed gravel bars in order to explain sedimentation patterns at
the reach, bar and bed-form scale. Field measurements of grain size displayed evidence for
a downstream fining trend, which was interrupted by coarse sediment from tributary inputs.
There was evidence of down-bar fining in grain size on several of the bars.
Using a 0.2 m moving window over non-detrended laser scan data, a positive relationship
between surface roughness and grain size was established, with an R2 value of 0.49. This
relationship was used to map grain size variations across the sampled bars. Grain size maps
provide an indication of sedimentation patterns, such as down-bar fining due to lateral
accretion, shadow zones or converging currents. However, there are issues of accuracy due
to errors and incomplete vegetation removal in the laser scan data, which result in the grain
size maps not being fully representative of field conditions.
10
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
A mobile application for CT-scan image segmentation benchmarking: Demo
and user-case study
Chang X1*, Delmas P1, Gimel’farb G1
1
Department of Computer Science, The University of Auckland, Room 384, Level 3, Science
Centre Building 303S, 38 Princes Street, Auckland, New Zealand
* Corresponding author
This speech gives an introduction to the analysis of soil CT-scan images assistance with
mobile devices. The outcome of the mobile application is aimed to provide a convenient tool
for soil scientists to create a CT-scan image segmentation benchmarking database.
Soil CT-scan images allow scientists to look inside the soil sample without disrupting its
structure. The segmented results are also used to measure the porosity and reconstruct the
3D model. In many cases, global thresholding techniques can be found by applying the
statistical modelling on the histogram, e.g. using ISODATA or Expectation Maximization. As
these techniques use specific optimizations developed on medical imaging, we introduce
modifications which fit the requirement of soil analysis.
Soil CT-scan images contain many regions, such as clay, sand, pores, water, organic matter
and minerals, and the respective area of each region may vary through different samples,
even when samples were collected from close locations. Due to such uncertainty, it is
challenging to apply a blind statistical method for the accurate segmentation of CT-scan data
of soil cores at large. Based on the manual thresholding results, we found that global
thresholds do exist but they cannot separate soil regions accurately enough.
A multiple-region Kriging method is introduced. The segmentation process is refined by
applying the conventional Kriging method on the outputs of multiple-class Expectation
Maximization algorithm. The method works on synthetic image with known number of
regions, but it is hard to assess its accuracy when there is no ground truth on the real soil
CT-scan images. Our mobile application provides a manual labelling tool to interactively
create such benchmark images, allowing experts to select seed points pertaining to different
regions of the soil. Through zooming operations, the accuracy is up to 1 pixel of the original
CT-scan image. After multiple points have been selected for each region, the application
computes the initial thresholds purely based on user’s inputs. Further refinement including
Kriging allows satisfactory segmentation of the image.
11
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Soil management, soil porosity, gas diffusion and nitrogen transformations
Thomas S1*, Balaine N2, Clough T2, Beare M1, Harrison-Kirk T1
1
Plant & Food Research, Private Bag 4704, Christchurch 8140, New Zealand
Department of Soil and Physical Sciences, Lincoln University, Lincoln 85084, New Zealand
* Corresponding author
2
Agricultural practices such as tillage, machinery loading and animal treading can dramatically
change soil structure. Compaction will normally reduce soil macroporosity while increasing
meso- and micro-porosity, altering the alignment, connectivity or tortuosity of those pores.
These changes directly affect solute and gas transport, soil aeration and redox status, in turn
affecting a range of soil biochemical and chemical reactions.
Several examples of the disruptive nature of soil management on soil gas processes and
how these affect soil nitrogen (N) losses are presented. Nitrogen is often the nutrient limiting
plant growth. However, N in the forms of nitrous oxide (N2O) and nitrate can be an important
environmental pollutant. Nitrogen transformations are strongly influenced by soil aeration and
redox state. Biological denitrification is a predominantly anaerobic process, whereas
nitrification occurs in more aerobic conditions. Other processes such as nitrifier-denitrification
and co-denitrification may occur across a range of aerobic conditions.
In a laboratory study, increasing soil compaction affected the rate and products of urinenitrogen transformations during three successive wetting-draining cycles. Macroporosity was
reduced with increasing compaction. This changed soil gas diffusivity with only moderate
increases in soil bulk density. Reduced aeration in the more compacted soil delayed
nitrification and altered the ratio of denitrification products (N2O and di-nitrogen) through the
wetting-draining cycles. In another laboratory study, a strong relationship between soil gas
diffusivity and N2O emissions through a range of levels of compaction was demonstrated.
Field plot experiments have demonstrated how compaction, urine treatment and their
interactions can lead to large N2O losses compared to non-compacted plots. Soil type can
strongly affect the total amount of N2O produced under similar compaction and urine
treatments. Nitrous oxide emissions were almost an order of magnitude greater from
compacted poorly drained soils compared to freely drained soils. The effect of compaction on
N2O emissions was as great as emissions following the deposition of high rates of N found in
cow urine patches (600 kg N/ha). Similarly in cropped fields N2O emissions from nonfertilised areas compacted by tractor wheels produced much greater emissions (2.5 to 3.4 kg
N2O-N/ha) than heavily fertilized potato plants (1.1 to 1.2 kg N2O-N/ha).
Results from these experiments demonstrate the importance of understanding how changes
in soil structure affect soil processes, in particular the regulation of processes that may have
detrimental effects on the environment. It also highlights the need to develop agricultural
management practices that can improve and maintain soil structure to reduce environmental
losses.
12
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Analysis of soil functions based on pore scale imaging: Computation of
unsaturated properties of a sandy soil
Oxarango L1*, Shiota E2, Ortega P1, Tinet A-J3, Delmas P4, Mukunoki T2
1
LTHE, University Grenoble Alpes, 38041 Grenoble, France
X-Earth center, University of Kumamoto, Japan
3
GeoRessources Lab, Université de Lorraine / CNRS / CREGU, Vandoeuvre-lès-Nancy, F54510, France
4
University of Auckland, NZ
* Corresponding author: [email protected]
2
Water flow and retention in the vadose zone constitute a key function in the general
environmental water cycle. They are generally studied at the core scale of the soil using two
relationships: (i) the retention curve describe the link between the average pressure (or
suction) and the volumetric water content and (ii) the relative permeability curve relates the
permeability and the volumetric water content. On a more fundamental point of view, these
processes involve capillary effects and fluid mechanics in the complex pore space geometry
of the soil. This study aims at using X-ray Computed Tomography (CT) for direct 3D imaging
of the 3 phases (solid, water, air) distribution inside a soil sample submitted to increasing
suction states. A custom hanging column apparatus is used to impose suction steps to a
sandy soil sample ( = 1.5cm, H = 1.5cm). The analysis of the phases geometrical structure
is currently achieved using a marker controlled watershed segmentation but still rises
perspectives for improvement. Work under progress concerns the direct quantification of
meniscus (air/water interface) surfaces that could provide an original comparison element
with the classical Laplace's theory of capillary equilibrium. The applicability of a graph cut
method based on a semi empirical extension of Cauchy-Crofton formula is discussed.
Finally, a local estimate of the retention curve is obtained from direct computation of the
water phase volumetric content. It compares satisfyingly with the classical approach based
on water mass balance. The relative permeability curve is estimated through direct numerical
simulation of the flow in the water phase using the Lattice Boltzmann Method (LBM). A
simplifying assumption is used considering that the water flow could occur inside the water
phase without changing the water phase distribution in the pore space. The resulting relative
permeability curve computed using CT images compares satisfyingly in term of order of
magnitude with the classical estimate based on grain size distribution for the saturated
permeability and the Van-Genuchten Mualem model for the relative permeability.
13
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
3D images for modelling behavior of natural porous media in geosciences
Tinet A-J 1*, Golfier F1, Kalo K1, Grgic D1, Giraud A.
GeoRessources Lab, Université de Lorraine / CNRS / CREGU, Vandoeuvre-lès-Nancy, F54510, France
* Corresponding author: [email protected]
Image acquisition and subsequent image analysis may be used to derive information such as
pore structure, phases’ distribution…Images may not only be used for observation or
morphological analysis but also as a base for pore scale modelling. Pore scale models may
be discriminated between models that use directly the imaged structure and models that use
information derived from this structure (topology, simplified shapes…). Each model and the
function it aims to represent may require different image analysis from segmentation to
shape recognition or subsampling. The presentation describes image analyses that may be
required depending on the model (Lattice Boltzmann, Computational Fluid Dynamics, Pore
Network, Micromechanics…). Extra attention is needed to accurately consider the simulated
volume compared to the REV which may require further treatment of the image such as
subsampling or downgrading. Two applications, micromechanic homogeneization of oolitic
calcite and multiphase flow and transport with Lattice Boltzmann models are presented as
illustration.
14
Second PROTINUS international workshop: « Image
characterization ? » 18-20 January, Auckland, New Zealand.
analysis
for
soil
structure
Reconstruction of 3-D Scenes Captured by UAV using a Stereo Camera System
Gee T1*, Delmas P1, Gimel’farb P1
1
Department of Computer Science, The University of Auckland, Room 384, Level 3, Science
Centre Building 303S, 38 Princes Street, Auckland, New Zealand
* Corresponding author
Unmanned Aerial Vehicles (UAV) are an emerging technology that has huge promise,
especially for problems that involve mapping or monitoring large tracts of terrain.
This talk focuses on a fast 3-D scene reconstruction algorithm used to build models of the
Whangateau (New Zealand) estuary floor from stereo UAV footage. The results of such
algorithms are of interest to biologists who aim to understand the spatial and temporal flux of
marine creatures dwelling on the estuary floor.
The algorithm itself is a type of SLAM (Simultaneous Localization and Mapping) algorithm,
which iteratively updates a growing model of the scene by locating the position of the camera
within the model and then merging the results. Processing starts with the first pair of stereo
images (of a sequence) captured by a set of cameras that have been calibrated. The
calibration parameters then allow for rectification of images, and after which, they are
prepared for stereo matching. After stereo matching, each scene is converted into a point
cloud that needs to be merged in order to construct a single model. The relative location of
each point cloud is approximated using feature point matching and a robust formulation
gradient descent. Merging of point clouds is then performed, with point cloud colours merged
with the aid of a blending algorithm, in order to reduce boundary effects.
The algorithm is shown to work, however it is susceptible to drift due to the fact that the
merging operation only considers subsequent pairs within the sequence. Thus as future
work, it is recommended that either the output of the algorithm is used to initialize a Bundle
Adjustment, or that the algorithm itself incorporates global constraints.
15