DEM Generation based on UAV Photogrammetry Data in Critical Areas
Giulia Sammartano and Antonia Spanò
(DAD) Department of Architecture and Design, Politecnico di Torino, Torino, Italy
Keywords: UAV Photogrammetry, DEM, DSM, GIS, Filtering, Landscape Heritage.
Abstract: Many Geomatics technologies based on the use of terrestrial and aerial sensor offer a significant support and
new potentialities in term of quickness, multi-scale precision, cost-cutting, and in short, sustainability. The
3D data and mapping products, above all the large-scale ones derived from aerial acquisitions (e.g. Unmanned
Aerial Vehicles, UAV) can be gradually adopted even when the context is not enough accessible or standard
airborne data does not fulfill the requested resolution and accuracy.
Starting from the availability of large scale UAV data, the paper is mostly purposed to examine the use of
tools aimed to generate DEM (Digital elevation model) from DSM (digital surface model) obtained from
UAV flights. In literatures many application concern the point cloud data generation from aerial
photogrammetry or airborne laser scanner. Several different filtering approaches and algorithms (filtering
point along density, direction, slope) are used to derive bare-Earth, but in the test case, the high level of detail
of objects, together with the complexity of high slope of ground impose some adaptation.
The test is included in a decision-making processes concerning the promotion of Alpine landscape leaded
through a project of sustainable mobility. Therefore the DEM generation is used to foresee a possible and
sustainable path of the railway rack, achieved by a simple multi-criteria analysis performed by Geographic
Information Systems (GIS) tools. In the end an important aspect of the test is the use of open source GIS tools
employed in the experience.
1 INTRODUCTION
Nowadays the documentation technologies based of
advanced Geomatics tools offer a significant support
in maps updating, in term of quickness, precision,
cost-cutting, and in short, sustainability.
The use of Unmanned Aerial Vehicles (UAV)
offers almost new potentialities with high detail
value, and related applications truly become
progressively affordable, even where the context is
not enough accessible for traditional terrestrial survey
techniques. (Aicardi et al., 2014)
A project of sustainable mobility in mountain
framework (a rack railway) have been analysed and
improved through the setting up of an integrated project
of 3D landscape documentation and modelling. This has
been achieved using aerial photogrammetry by drone and
3D data treatment and spatial analysis in GIS. In
landscape documentation, the use of UAV and integrated
sensors, together with the application of more traditional
Lidar terrestrial and aerial techniques, perform very
interesting results in the context of large-scale mapping
and in terms and effectiveness.
In case of complex documentation applied at
specific valuable zones of the built and natural
heritage, that are located often in arduous areas very
scarcely accessible, the previous issues lead us to
choose specific survey tools able to reach great detail
and at the same time, a relatively large area of
coverage. (Boccardo et al., 2015).
2 DEM FROM UAV
The UAV approach can be useful to produce spatial-
temporal high-resolution models, in competitive period
and resources, that providing useful 3D data for many GIS
monitoring applications, as georeferenced information at
large-scale derived from orthoimages and DSM (Yastıklı
et al., 2013).
The main application of UAV survey producing
spatial data for GIS modelling and analysis are the
territorial, geological, urbanistic, agricultural and forestry
ones (Perko et al., 2015; Höfle et al., 2013; Susaki, 2012;
Grenzdörffer et al., 2008). Even in the architectural and
archaeological contexts the use of UAV become
increasingly important and common, adding to essential
to achieve accurate metric documentation, integrate
92
Sammartano, G. and Spanò, A.
DEM Generation based on UAV Photogrammetry Data in Critical Areas.
In Proceedings of the 2nd International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2016), pages 92-98
ISBN: 978-989-758-188-5
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
different source geospatial data, as well as categorize and
store spatial-temporal information. (Themistocleous et al.,
2015
; Rinaudo et al., 2012)
For investigation purposes about land forms the
use of UAV acquisitions enables to obtain high
detailed morphological data of the area, that can be
processed in order to generate a DSM, comparable to
the Lidar ones; with these results we can proceed in
many cases to update effectively traditional
cartography and numerical data. However, the 3D
information regarding the earth surface need to be
processed for the derivation of a useful DEM, without
the interference caused by vegetation cover and
buildings. In literatures many application in these
direction concern point cloud data from aerial
photogrammetry or airborne laser scanning and
different filtering approaches and algorithms.
(Hosseini et al., 2014; Korzeniowska et al., 2011,
Chen et al., 2007; Masaharua et.al., 2002).
In many cases, the bare-Earth extraction can be
obtained with algorithms of point clouds classification
and segmentation, by filtering point along density,
direction, slope, etc. (Pfeifer, 2008)
This takes moderately a large amount of time of
human action in modelling systematic errors, filtering,
feature extraction and thinning, up to 60%-80% of the
processing time (Sithole, 2004).
This becomes more complex in the case of areas
where the ground is not flat and common algorithm of
analysis in high density urban, peri-urban or forestry areas
(Perko, 2015. Susaki, 2012, Korzeniowska, 2011), that
recognize objects according to their variation of high and
density from ground, as an almost plane level do not work
effectively. In past year, some procedures have been
developed for slope land applications (Ismail, 2015.
Hosseini, 2014), and adaptive algorithms where the
threshold varies with respect to the slope of the terrain
have been implemented (from the proposed one by
Sithole, 2001).
3 THE EXPERIENCE ON ALPINE
SLOPES
The overall experience of high scale mapping of the
hamlets in the Castelmagno area (Piedmont, Italy)
had numerous purposes, combining terrestrial
scanning and low aerial acquisition with the aim to
achieve multi-scale models. It’s not objective of this
paper the description of used instruments and
techniques, since we are going to evaluate the use and
processing of DSM derived from the UAV flights.
The DSM and orthophoto generation have been
accomplished by Structure from Motion (SfM)
combining photogrammetry and computer vision
methods. In spite of the availability of two flights, the
first one from an hexacopter flying at the height of
about 70 m. and a second by a fix wing Ebee drone
flying at the height of 120m, we are going to consider
only the second source. The next results are the
starting point of the present work. Covered area: 2.43
Km
2,
ground resolution: 0.11 m/pix, camera stations: 544,
Tie-points: 1044668 GCPs residual: 2.5 pixel, CPs
residual: 3 pixel
3.1 The Need of Large Scale Map
Updating
The placement of a railway rack as a tool of local
development has been designed to provide a large
fruition, satisfying a tourist need and a private use.
The feasibility study expects that a path
connecting three hamlets could promise a process of
repopulation of the area in order to re-establish the
main economic activity as desired.
So the railway rack has to cross in a transverse
direction the hydro-geological basin of Valliera,
which is a tributary of the Grana River. The departure
station is designed to be placed in the Colletto village
(1277 m above sea level), the path has to descend fast
toward Croce (1179 m a.s.l) crossing the riverbed and
climb the slope to Campofei (1542 m a.s.l.)
Figure 1: UAV orthophoto of the test area; the arrows show
three villages to be connected by the railway rack.
First of all, the challenge to foresee a possible and
sustainable path of the railway rack, need a very
detailed DTM (Digital terrain Model), more accurate
than the regional 1:10000 scale DTM. Next
paragraphs are devoted to describe a DTM generation
from UAV DSM, using open source techniques. Then
a best routing search is presented in the last
paragraph.
DEM Generation based on UAV Photogrammetry Data in Critical Areas
93
3.2 DSM to DEM Conversion
The aim of the first step of the test is the use of UAV data
to obtain a morphological model of the project area,
starting from the DSM, analysing and eliminating objects
surfaces, and finally generating a DEM. The difficulties
are attributed to the complex mountain context in with a
forestry covering very diffuse was in many zones very
thick. Furthermore, the steep gradient of the mountain
slope added a robust bond to the raster analysis based on
simple available algorithms distance range based.
Figure 2: The DSM of the test area, compared with the
orthoimage from the drone flight: it is distinct the assorted
presence of vegetation cover in a very slope terrain.
3.2.1 Managing UAV Data in DSM
Filtering: Different Approaches
The operational approach to derive a DEM from the
DSM calculated from the UAV images was based on
processing raster data by filtering algorithms.
The test was planned on the Open-Source platform
QGIS (QGIS 2.10 Pisa, http://www.qgis.org/en).
The software offers many simplified and advanced
processing tools, also with algorithms extension from
GRASS Gis and SAGA Gis. In the test both of them
have been used to setting up the DSM analysis.
The two raster approach proposed started from a
base-objective, that is to recognise and eliminate non-
coherent objects like buildings and wooden covering on
the soil. It can be achieved with direct application of
Raster filters, but not easily suitable because of the slope
or, indirectly operating a multiple-step workflow of
raster calculation and modelling (Cimmery, 2010)
DTM Filter Slope-based on SAGA GIS, allows
recognizing object according to input data of search radius
and terrain approx. slope. After several attempts, the best
combination of values for the area have been: terrain
slope: 30%. Search radius: 20. The output is a double
divided raster data, with detection of Bare Earth (figure
3). This result is not enough appreciate, despite high input
values, because of the internal area of removed object
profiles that are still tree and building mass.
Morphological Filter on SAGA GIS, manages to
smooth the raster DSM according to a search mode
(circle or square) and according to 4 algorithms
(dilation, erosion, closing, opening). The best fitting
algorithm to model the area have been the erosion
filter on circle mode with radius 50.
Gaussian Filter on SAGA GIS allows a filtering of
the raster data according to the Gaussian histogram,
imputing the Standard deviation.
Figure 3: Example of Removed objects and Bare-Earth, the
output from a DTM filter slope-based.
Figure 4: (Left) One of morphological filter outputs applied
in test area with erosion 50 rad in circle searching mode
(right) Gaussian filter STDEV 20, radius 70.
3.2.2 Helpful Approach for DEM
Conversion
The proposed workflow tries to integrate some
procedures of raster management in order to identify
and remove trees and building, understood as
discontinuity in the trend of mountainous terrain
sloping. In this sense the Ruggedness Index have been
applied to the threshold suitable to maximize the
finding of objects on the area. Then a buffer area from
3 to 5 meters have been identified, that tries to include
all the items to be deleted from the model.
Afterwards a cutting mask have been created to
remove from the DSM incongruous elements. After
this step, the following actions have been directed to
fill hole after the cut and compare and assess the result
of final DEM with the original DSM. (Fig.10). Even
if some few residuals of final DEM points close to
trees and buildings are still high, we can consider the
valuable result of the surface interpolation and objects
removing, taking into account the critical situation of
the soil morphology.
GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management
94
Figure 5: Applying the Terrain Ruggedness Index
algorithm to de DSM: best-input value 0.05.
Reclassification of the raster with binary values (0-1) to
perform buffer analysis on a unique value.
Figure 6: (Left) Creating a buffer area around objects by
using the Proximity (Raster Distance) analysis tool. (Right)
Identification of trees and buildings and vectorialization of
raster in shape file layer.
Figure 7: Recognized objects after raster cut on the DSM.
Figure 8: The closing gaps filters on SAGA GIS according
to tension threshold 0.1.
Figure 9: Comparison between the DSM generated from
UAV image processing and the final DEM obtained from
GIS raster analysis and filtering.
Figure 10: Residuals of DSM/DEM from GCP. The
elaborated DEM, differs from DSM mainly on rich
vegetation and built-up areas.
Figure 11: DEM residuals of on each GCP measured by GPS
survey, clustered by their position in the area. Higher residuals
between DEM and GCP are near buildings and vegetation.
3.3 Path Hypothesis
3.3.1 First Shortest Path Proposal
Based on an edited and evaluated DEM, the first aim
of the has been the assessment of suitability of the
shortest distance route, that correspond in a first
approximation to the cheapest.
DEM Generation based on UAV Photogrammetry Data in Critical Areas
95
Figure 12: Slope map in red range color with firstly designed
route of railway rack. (green). (right) Comparison of elevation
differences (black) and slope values (green) of the straight path
connecting the villages.
Safety first this straight path hypothesis has been
compared with the regional map of landslide risk.
Fortunately, the presence of a ridge submit to risk a
face immediately next to slope of designed path, in
the west direction.
The slope map generation, derived from the
available high scale and accurate DEM, has brighten
the presence of very precipitous slopes, in fact very
often the hamlets have a medioeval or XVI- XVIIth
foundation, so they are grasped to stone walls that are
typical of the geo-morphological characterisation of
the Valle Grana.
3.3.2 A Simple Multicriteria Analysis to
Search a More Suitable Path
The abilities offered by the continuously expanding GIS
technology and by the increasing availability of digital
georeferenced data have been highly develop the use of
multi-criteria spatial analysis to support land planning.
(Botequilha-Leitão, Ahern, 2002).
Close to the riverbed, the slope reach to 43% and even
worse many railway rack system are not able to divert the
slope.
Indeed some innovative systems in use in alpine
environment are able to reach very high slopes, even
40%. (Regis, 2015). The graph of figure 12, show the
terrain profile highlighting the level differences
among villages, contemporary with the slope values
that reaches 43% close to the river.
Moreover, the GIS-Multi-criteria Decision
Analysis is largely used to support best location
analyses in very many fields of application. In this
paper, we are going to use an example of a so called
weighted overlay performed by raster grids, with the
purpose of providing a test for the achieved UAV
DEM.
Surely, the stronger constrain for searching a
sustainable and best route is the slope, and we know
that ordinary types of railway rack systems are able
to face slopes around 20%, but some innovative types
provide equipment with a successful power facing
slopes higher than 40%. This last type will be taken
into account for the best route assessment.
The slope raster map has been reclassified
imposing four class of slope: from 0% to 20, from
21% to 35% , from 35% to 40%, and more than 40%.
(fig. 13). The last class will be considered highly
adverse.
The shortest route cross the valley using almost
the steepest path, and the change of slope direction
close to the riverbed is particularly difficult. Another
considered parameter is the inclination the railway
rack route chasing contours direction; the DEM grid
has been reclassified in four classes, selecting the
breaks value in order to obtain an area near the river
to be excluded, or considered as highly adverse area.
Figure 13: a. four classes of slopes grid map. b. five classes of elevation interval c. Parcel whose owner participate to the railway rack
project.
GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management
96
(Orange in fig. 13b). The last parameter involved in
the process are parcels whose owner has agreed and
participated to the projects or not.
In the weighted overlay analysis, the three maps
have been different level of influence. The slope
classes had been an influence of 70%, the grid map
connected with contours had a 20% influence, and the
last one has been considered with low influence
(since the administrative institution could give some
benefits for the project acceptance).
The final result of weighted overlay is shown in
figure 14, where green pixels show a highly suitable
area to accommodate the railway rack, the dark
yellow area are on average suitable, and the last class,
in red pixel correspond to unfavorable area for the
railway rack.
In Figure 14 the shortest and steepest path (red)
has been compared with a manually traced route
(green), avoiding unfavorable areas.
Figure 14: The grid maps visualize by different color the different
suitability of areas to host the railway rack passage.
4 CONCLUSIONS
The increasingly studies as well as diversified
applications of UAV photogrammetry survey make
consider it a very suitable method for high scale quick
and low cost mapping. Many processing techniques
and platforms can manage today DSM derived from
UAV processed images; and we are witnessing to a
strong effort in order to adapt strengthened tools
manage this kind of data in critical areas.
This paper is aimed to prove that the use open
source tools to achieve the analyses and can be a
significant step toward that knowledge circulation
and sharing about geospatial data overall. Surely,
some more enhancements need to be implemented in
terms of big objects filtering tools. Some uses
promise to offer further developments, especially in
benefit that the GIS-based landscape modeling can
offer to the analysis and decision-making phases in
the ever more topical background of land use
planning and heritage.
ACKNOWLEDGEMENTS
The UAV flights have been performed by the DiRecT
Team (Disaster Recovery Team), involving Aicardi, I..
Chiabrando, F. Donadio, E. Lingua, A. Maschio, P.
Noardo, G. Sammartano, F. Spanò.
REFERENCES
Boccardo, P., Chiabrando, F., Giulio Tonolo F., Lingua, A.,
Dutto, F., 2015, UAV Deployment Exercise for Mapping
Purposes: Evaluation of Emergency Response
Applications. In: Sensors, vol. 15(7), pp. 15717-15737.
Ismail, Z., Abdul Rahman, M. Z., Mohd Salleh, M. R., Busu,
I., Amerudin, S., Wan Kadir, W. H. 2015. An improved
progressive morphological filtering algorithm based on
spatially-distributed slope value over tropical vegetated
regions. Jurnal Teknologi, 77(26), 87-93.
Regis D, 2015, Progetto di infrastrutture e territorio alpino
contemporaneo, ArchAlp, 9.
R. Perko, H. Raggam, KH. Gutjahr, M. Schardt, 2015.
Advanced DTM generation from very high-resolution
satellite stereo images, ISPRS Annals, Volume II-3/W4.
K. Themistocleous, A. Agapiou, B. Cuca, D.G. Hadjimitsis,
2015. Unmanned aerial systems and spectroscopy for
remote sensing applications in archaeology, ISPRS
Archives, Volume XL-7/W3.
Aicardi I., Boccardo P., Chiabrando F., Facello A., Gnavi L.,
Lingua A., Pasquale F., Maschio P., Spanò A., 2014, A
didactic project for landscape heritage mapping in post-
disaster management., in Applied Geomatics.
S. A. Hosseini, H. Arefi, Z. Gharib, 2014. Filtering of Lidar
point cloud using a strip-based algorithm in residential
mountainous areas, ISPRS Archives, Volume XL-2/W3.
B. Höfle, L. Griesbaum, M. Forbriger, 2013. GIS-Based
Detection of Gullies in Terrestrial LiDAR Data of the
Cerro Llamoca Peatland (Peru), Remote Sensing, 5,
5851-5870.
N. Yastıklı, I. Bagci, C. Beser, 2013. The processing of image
data collected by light UAV systems for GIS data capture
and updating, ISPRS Archives, Volume XL-7/W2, 2013
ISPRS2013-SSG.
F. Rinaudo, F. Chiabrando, A. Lingua, A.T. Spanò. 2012.
Archaeological site monitoring: UAV photogrammetry
can be an answer. In: ISPRS Archives, vol. XXXIX n. B5,
pp. 583-588.
J. Susaki, 2012. Adaptive Slope Filtering of Airborne
LiDAR Data in Urban Areas for Digital Terrain Model
(DTM) Generation, Remote Sensing 2012, 4(6), 1804-
1819.
DEM Generation based on UAV Photogrammetry Data in Critical Areas
97
K. Korzeniowska, M. Łackam, 2011. Generating DEM
from Lidar data – comparison of available software
tools. Archives of Photogrammetry, Cartography and
Remote Sensing, Vol. 22, 2011, pp. 271-284.
V. Cimmery 2010. User Guide for SAGA (version 2.0.5),
in: http://sourceforge.net (Accessed 14 January 2015).
G. J. Grenzdörffer, A. Engel, B. Teichert, 2008. The
photogrammetric potential of low-cost UAVs in forestry
and agriculture. ISPRS Archives. Vol. XXXVII. Part
B1.
N. Pfeifer et all. 2008. DSM / DTM Filtering, International
School on Lidar Technology 2008 IIT Kanpur, India
(from: http://www.iitk.ac.in/).
Q. Chen, P. Gong, D. Baldocchi, G.Xie, 2007. Filtering
airborne Laser scanning data with morphological
methods, Photogrammetric Engineering & Remote
Sensing Vol. 73, No. 2, February 2007, pp. 175–185.
G. Sithole, G. Vosselman, 2004. Experimental comparison
of filter algorithms for bare-Earth extraction from
airborne laser scanning point clouds, ISPRS Journal of
Photogrammetry & Remote Sensing 59 (2004) 85– 101.
Botequilha-Leitão, A., Ahern, J., 2002, Applying landscape
concepts and metrics in sustainable landscape
planning. Landscape and Urban Planning, 59.
H. Masaharu, K. Ohtsubo, 2002. A filtering method of
airborne Laser scanner data for complex terrain,
ISPRS Archives, Commission III, 34 (3B): 165-169.
G. Sithole, 2001. Filtering of laser altimetry data using a
slope adaptive filter, ISPRS Archives, Volume
XXXIV-3/W4.
GISTAM 2016 - 2nd International Conference on Geographical Information Systems Theory, Applications and Management
98