aerial oblique lidar-photogrammetry mapping for geology… · 2011-12-09 · 38th zlg eth (cas)...

38th ZLG ETH (CAS) Short Course Remote Sensing and New Monitoring Techniques in Applied Geology – 24-28.10.2011 - ETHZ 1

AERIAL OBLIQUE LIDAR-PHOTOGRAMMETRY MAPPING FOR GEOLOGY: INTEREST OF OBLIQUE DATA ACQUISITION, APPLICATIONS IN GEOLOGY (STRUCTURAL MODELING,

MASS BALANCE, MONITORING OF MOVING TERRAIN), LIMITATIONS

Julien Vallet, Barbara Haebler Helimap System SA

Abstract During the last decades, the arrival of new mapping technologies such as Lidar brought new possibilities for the mapping of complex or inaccessible areas. The development at EPFL of a handheld helicopter based mapping system relying on Lidar-photogrammetry issued on a commercial production system: Helimap system®. The integration of the Lidar, Direct georeferencing techniques and digital photogrammetry into a handheld unit permits to fit the topography with a flexible oblique mapping and provides high accuracy and resolution Digital Terrain Model and orthoimages. Then the use of the helicopter allows mapping extremely complex terrain such as high cliffs or steep slopes. After a summary of theoretical elements, application to geosciences (geology, geomorphology, natural hazards) is illustrated through practical experiences.

1. Introduction

Airborne laser scanning is now a recognized technique to provide accurate 3D data of our environment. This technique released in the mid 1990’s proved quickly its great performance for high resolution mapping of remote areas. The combination with digital photogrammetry compensated the drawbacks of the laser technique for thematic classification of land cover (using orthoimage) or accurate linear objects modeling. In both cases, the sensors are fixed to the aircraft and look downward (“nadir”). This conventional configuration suits well for smooth terrain or large area mapping but it reaches its limits when facing to complex terrain with steep slopes or cliffs. In a conventional way, the slope involves a non uniform scale of imagery and thus a non uniform mapping accuracy (e.g. Kraus 1998). Regarding the LiDAR, the slope acts on the footprint size of the laser beam and on its incidence angle. Thus, the bigger is footprint, the more uncertain is the distance measurement due to averaging effect inside the footprint. Moreover, all the planimetric errors induce altimetric errors while the slope increases (Favey, 2001). On flat areas, for a given planimetric/altimeric accuracy of 20/10cm, the altimetry decreases to ~30cm for 45° slopes and can be significantly more for cliff or vertical areas (Vallet, 2002). Moreover, the vertical configuration does not permit to map overhangs or caves. An oblique acquisition allows getting more homogenous laser and image data because the aiming axis is close to the surface normal and then any slope is considered as flat. The second drawback of a fixed system is generally the complexity of the system setup or the availability of the carrier if the system is permanently mounted, which prevents from using it in an emergency way at reasonable cost. The market of small area and high accuracy mapping did not seem to hit sensor providers and no commercial system permitting fast setup and oblique mapping is sold. It is in this spirit that a handheld helicopter based mapping system, combining photogrammetry and LiDAR is born. (Skaloud et al., 2005). The purpose of this paper is first to summarize theoretical concept of LiDAR-Photogrammetric mapping and then to present the original and unique concept of handheld oblique mapping and its mapping performance. As originally designed for avalanche monitoring in the Alps, we will show


through several experiences the real potential of such light and flexible system in the field of geosciences and mountain mapping.

2. Theoretical concept of LiDAR-Photogrammetry combination Photogrammetry and LiDAR are two complementary mapping techniques and the integration of both into a single data acquisition provides the best of each one:

♦ Automation and high density of LiDAR measurement ♦ Measurements through vegetation coverage with LiDAR ♦ Radiometric information of image: interpretation of the scene with a higher resolution than

LiDAR In the following chapters below, a brief theoretical summary is done. For more details, refer to Vosselman et al. (2010) for LiDAR and to the manual of photogrammetry (Kraus et al., 1996).

2.1 LiDAR

2.1.1. Principle LiDAR technology relies on electronic distance measurement (EDM). The distance is estimated by measuring the time of flight (TOF) of an electromagnetic wave between the emitting source and the reflection on the target object according to the following relation:

2

tcd

⋅=

with c as speed of light and t the time of flight (go and back).

2.1.2. LiDAR characteristics

Airborne mapping LiDAR are composed of an emitting light source (Laser), a detection sensor (photodiode), a scanning device (fig.1), and a georeferencing system that permits to position and orient the LiDAR in the object space. The main characteristics of a LIDAR unit are:

♦ Pulse Repetition Rate (PRR): it represents the number of Laser pulses sent by second. The PRR varies from 10 to 400 kHz and depending on the scanning technique, it can differs from the real measurement rate.

♦ Scanning mode: generally done with a rotating or oscillating mirror. Main modes are illustrated in the figure 1.

Figure 1: Illustration of the main scanning methods of LiDAR. From left to right, the rotating mirror (lines), the oscillating mirror (sawtooth) and the elliptic scanning. The scanning is only 2D (according to one line pattern) and it is the motion of the aircraft which permit to scan the object.

Field of View FOV FOV FOV

Flight �


♦ Scanning rate: it is the number of scan per second. The range is between 10 and 100 Hz. ♦ Beam divergence: the beam divergence represents the aperture of the laser beam. This beam

aperture leads to a laser footprint on the target. Higher is the divergence, larger is the footprint. The footprint size has an important influence on accuracy especially in steep slopes.

♦ Wavelength: the intensity of the laser reflection is influenced by the wavelength. For example, reflection on snow is ~80% until 1100 nm and is only 15% at 1200 nm.

♦ Distance range: As the TOF of a pulse is measured, the limitation of the distance measurement is generally the speed of light for a given PRR since pulse is not sent before the previous returns. Here we have the LiDAR paradox: higher you are, lower the PRR should be, and in fact it is the contrary that we would need. To face this physical fact, some new techniques has been developed to detect pulses unambiguously before the previous returns. It is called Multi-pulse in the air (MPiA).

♦ Multiple returns – echo: this is the most interesting aspect of the LiDAR. Its capability to record several returns for one emitted pulse (fig.2). This permit to measure at the same time ground and ground cover or aerial objects (forest canopy, cables, wire…)

Figure 2 : Illustration of the multi echo, and its signal shape (amplitude) vs. footprint (Courtesy of Riegl)

2.1.3 Georeferencing To obtain the LiDAR point coordinate in the ground coordinate system, it is necessary to know the range measurements between the LiDAR and the object, the scan angle of the mirror, the position of the LiDAR origin and its orientation (fig. 3). The position and orientation are given by a direct georeferencing system composed of an inertial measurement unit (IMU) and a GPS-GNSS receiver. The IMU records the acceleration and rotation rate at few hundreds hertz while the GPS records the position and velocity of the LiDAR. Both measurements are integrated to provide position and orientation angle for each LiDAR measurement. The georeferencing equation of LiDAR is:

( )( )PPALSIMU

IMUALS

EIMU

EIMU

EP dReRRXX ⋅+⋅⋅+= α

with: EPX : Position of point P in the object frame E (ground)


EIMUX : Position of IMU origin in E (indirectly provided by GPS antenna position, i.e.

GPSIMU

EIMU

EGPS

EIMU eRXX ⋅+= )

EIMUR : Rotation matrix between IMU (body) and object frame (E) � angular measurement from

GPS-IMU integration : roll, pitch, heading. IMUALSR : Boresight matrix (misalignment between frame axis due to assembly) between IMU and LiDAR

(ALS) ALSIMUe : Lever arm IMU - LiDAR

GPSIMUe : Lever arm IMU – GPS antenna

R(αP) : laser beam direction vector in the LiDAR frame (function of α)

dp : range measurement of LiDAR

Figure 3 : Direct georeferencing of LiDAR

2.1.4. Error sources

Error sources in LiDAR mapping are numerous (Schenk, 2001) can be separated in direct and indirect component. Direct sources include errors of system sensors (position, orientation, range…) while indirect sources depend on object nature (material, rugosity) combined with incidence angle of the beam. Those errors are more difficult to quantify.

Error budget can be split in:

♦ Sensor position error: Trajectory position accuracy is generally within a range of 5-15cm. This error affects the points with a shift dx, dy, dz, and is evolving with time with satellite visibility and constellation geometry. It is generally the main error component for low altitude flight. (<200 m a.g.l) (Glennie, 2007)

GPS ANTENNA XE

GPS z

y

κ φ

ω

LiDAR

XEIMU

X E

R IMU ALS

P Objet E

IMU Body Frame

Y E

ZE roll

pitch azimuth

y

z

x

ααααP

XEP

RE IMU

e ALS IMU

d P

x


♦ Range measurement error: It is a function of the signal to noise ratio (SNR). It depends on type of soil/material and on the atmosphere along the laser path. It is generally given to a range of 20-30 mm but it is often better on hard smooth surface at low flying height (~10 mm).

♦ System orientation error (IMU, mirror encoder): this error increase linearly with distance (fig. 4). It is given by the following relation:

)cos( ALSimuxy edE α⋅⋅≈ (E.5)

[ ]pedE ALSALSimuz tancossin ⋅+⋅⋅≈ αα (E.6)

On flat terrain, planimetric component is the major one. This is accentuated by the fact that the heading accuracy is generally 2-3 times worse than the roll and pitch accuracy.

Figure 4 : Angular error influence on

ground measurement error

♦ Beam divergence δδδδ : the range measurement results of the distance integration on the entire laser footprint. Bigger is the footprint, noisier is the range. The increasing of the size and the deformation of the footprint on oblique incidence degrade the range accuracy. Reflectivity heterogeneity is also an important error source is the footprint is large (fig. 5).

Figure 5 : Example of divergence

influence with a differential reflectivity

2.1.5. LiDAR accuracy and performance Point cloud accuracy depends on angular accuracy of orientation, flying height, beam divergence and position accuracy. Angular accuracy is conditioned by the IMU grade (tactical or navigation). For high flying height (> ~500m), navigation grade IMU are highly recommended to preserve a good heading accuracy. The figure 6 shows the theoretical accuracy of laser point as a function of flying height, IMU type and incidence angle in the swath (nadir and edge) for a given sensor position accuracy of 3-5cm on X Y and Z and for a flat terrain of uniform reflectivity.

αALS eimu

d

p Ez Exy

Distance mesurée Angle mesuré

measured real point

paint


Précision d'un point au sol vs. type d'IMU (N=Navig ation T=Tactique) sur sol plat

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

0 250 500 750 1000 1250 1500 1750 2000 2250 2500

Hauteur sur sol [m]

Pré

cisi

on (

1σσ σσ

) [m

]

Sigma XY Nadir N Sigma Z nadir N

Sigma XY FOV 60° N Sigma Z FOV 60° N

Sigma XY Nadir T Sigma Z Nadir T

Sigma XY FOV 60° T Sigma Z FOV 60° T

Figure 6 : Theoretical accuracy of ground point vs. flying height, IMU grade and FOV on flat terrain.

2.2 Photogrammetry integrated to LiDAR

The integration of photogrammetric data to LiDAR system benefits from the direct georeferencing sensors. It simplifies the process of aerial-triangulation (AT). It also provides high resolution images which complete the point cloud data set. Images can thus be used to:

♦ create ortho-rectified images ♦ make stereo plotting of linear or punctual specific elements ♦ make interpretation of the surface (digitizing, classification) ♦ help for LiDAR point classification ♦ extract dense point cloud ♦ texture digital terrain model (textured mesh)

As direct georeferencing (DG) provides the camera orientation at the time of the shot, aerial-triangulation could be avoided. But for accurate measurements, the use of the DG orientation could be problematic: undistributed residual parallaxes prevent from doing accurate stereo plotting, internal parameters of the camera can not be controlled without AT. A simplified aerial triangulation is generally recommended to remove those effects and to check the camera parameter stability (Vallet, 2009). 3. Handheld System design for oblique mapping

3.1. Context

The development of Helimap System started in 1998 in the photogrammetric and geodesy laboratories of Swiss Federal Institute of Technology of Lausanne (EPFL), in collaboration with the Swiss Federal Institute for Snow and Avalanche Research (WSL-SLF), to measure snow volume on avalanche sites (Vallet, 2002; Sovilla et al. 2006). Needs were a high mapping accuracy (~10cm) in complex terrain, a fast deployment to react in real time with weather condition and an affordable cost for small surfaces (< 2’000 ha). The project initiated with photogrammetry based on simple handheld camera and enhanced with Direct Georeferencing (DG) sensors, a digital camera and a laser scanner. Since 2005, the system is


operated by Helimap System SA and is still evolving with sensor technology. Among 300 flights, about the half were realized for mountain areas and natural hazard purpose.

3.2 Problematic of complex terrain

Complex terrain such as alpine regions presents steep slopes and cliff with overhanging areas. Acquiring LiDAR / Photogrammetric data with a conventional nadir view in steep areas presents several drawbacks: the density / resolution and the accuracy are heterogeneous and overhanging areas are generally not covered (Vallet, 2009). It is necessary to look at the ground perpendicularly to the slope (fig. 7) to avoid the slope effect. Oblique data acquisition can only be performed by using a helicopter since it is possible to fly parallel to the slope.

Figure 7: Nadir view acquisition (left) vs. Oblique (right) of steep/cliff area. The rotation of the sensor head to look perpendicularly the terrain allows getting homogenous accuracy and density. Overhangs are also covered. To be able to fit the data acquisition to the slope, the system concept must rely on the following key points:

♦ The high flexibility of the helicopter which permits to rove easily in the most complex terrain ♦ A constant accuracy and information density whatever the slope: Operating the system as

“handheld” provides a full freedom of motion and rotation to fit to the slope. Thus the system can be used either in vertical or oblique configuration. (fig. 8).

Figure 8: Vertical (nadir) and oblique configuration

Degraded accuracy No data


♦ To allow inaccessible area mapping. Thanks to DG techniques (GPS-IMU), it is possible to have a mapping accuracy better than 10cm without any ground control points (GCP) in the field.

♦ Time flexibility and fast deployment. Independent from helicopter type, the system can be operational in few hours.

3.3. Handheld sensor head components

To meet the needs, the system is designed with a modular assembly that integrates 4 sensors into a unique rigid bloc: a digital camera of 50 MPix with a Field of View (FOV) of 57°, a carrier phase dual frequency GNSS receiver, an inertial measurement unit (IMU) and a laser scanner with a measurement rate of 10’000 to 190'000 points/sec. The maximum range of the scanner is about 1’000 m and the FOV is similar to the camera (60°). The setup time is less than one hour on most helicopters.

Figure 9: components and system assembly (left) and system operation (right)

3.4. Mapping accuracy

The theoretical chart of figure 6 has been confirmed with ground control measurements and internal analysis within the point cloud. At a flight height of ~300m, the absolute and relative accuracy of the handheld sensor is shown on the figure 10 for both LiDAR and photogrammetric component (Skaloud et al. 2005).

0

0.05

0.1

0.15

Relative Absolute

LiDAR mapping accuracy 300m/agl

Road Snow Grass Rough

Photogrammetric accuracy [m] 280m/agl

0

0.05

0.1

0.15

X Y Z

AT AT-GPS AT-DG

Figure 10: Absolute and relative altimetric mapping accuracy of point cloud in function of ground rugosity (left). Mapping accuracy of photogrammetric data at 300m a.g.l. using aerial-triangulation or direct-georeferencing (right).


4. Practical experiences in geosciences and natural hazard mapping

Initially designed for snow volume measurements on steep slopes, the application field widened considerably to other domains such as corridor mapping (river, coast, road, and power lines). Nevertheless, the original use of a handheld system represents a unique advantage for complex terrain mapping. Geology and its related applications such as natural hazard studies or monitoring and geomorphology require 3D data in remote areas and steep terrain. The applications and limitations of oblique LiDAR-Photogrammetric mapping are described below.

4.1. Volume/mass balance Comparison of data series is a major part of the monitoring mass motion and transport. The high density and accurate Laser DTM’s involves precise volume measurements when comparing 2 or several episodes. This can be applied to any mass transport or surface evolution such as avalanche, landslide, debris flow or glacier mass balance, snow melt. Several flights were conducted on avalanche study site of the SLF, for ice and rockfalls (Fisher et al. 2011) (Monte Rosa E face, Randa, Gondo, Vercors), for snow/glacier melt following (Arolla, Tsanfleuron), or for permafrost area deformation monitoring (Dorfbach). The figures 11 show several volume and differential maps of such application.

Figure 11: illustration of differential computation of LiDAR DTMs. a. Rockfall in Vercors (Chamousset). Differential profile of the fallen flake. The area is small (30x40m) and was only accessible using helicopter.

10m

a. b.

c.

d.


b. Glacier d’Arolla monitoring. Height difference in 4 years with a loss volume of 28.106 m3. The glacier catchment is about 2’000 ha and is covered with a mix acquisition (oblique and vertical) in 6h. c. Major landslide in Martinique (Precheur River). Height map of the starting zone. The starting zone was clear but it was also possible to detect other minor landslide. d. Dorfbach monitoring in Randa. Permafrost melting release material and create debris flow. The area is flown 2-3 times a year to compute surface evolution. In the top of the area, boulder cluster is moving few meters per year while at the bottom, the debris flow creates erosion at the bridge.

4.2. Geomorphology, structural modeling The high resolution of laser data shows morphologic details that were invisible with the use of conventional photogrammetry because the cost to get such density was too high. Then, those high resolution DTM’s are particularly useful for geologist either in terms of monitoring of instable regions or for geo-structural analysis. The figure 12 depicts a landslide area near the Lac of Monteynard (France) studied by the CISM (University of Savoie). A local height variability analysis permits to identify waves characterizing the instability of the flank.

Figure 12: Landslide of Avignonet-Harmalière (Knieb et al., 2009). The left figure illustrates the local height variability. These analyses show the micro or local relief. The waves in green and brown depict the sine shape of the landslide. Perpendicular lines (brown) to the slope represent the pattern of the laser measurements. The right image shows the shaded relief of the area color-coded with orientation. High resolution DTM’s permit to identify surface deformation, faults or to model precisely outcrops. Combined to geophysical data (Ground penetrating radar…), it is possible to model the 3D structure of the subterranean layers. Several oblique projects were lead to bring high resolution data in vertical areas. Then every detail even in overhanging zone is captured. In this case, the LiDAR data serves as 3D base but the high resolution images are draped to the DTM mesh to create a 3D textured model. Thus, it is possible to digitize in 3D layers, cracks directly on the texture with a resolution of 3-10cm pixel. The figures 7 represent DTM’s of cliffs acquired for geo-morphological studies and structural


modeling (Svalbard, Congo, South Africa, Dolomite) that will be used to validate a 3D geo-structural model (Buckley et al. 2008).

Figure 13: 3D model of the Mediumfjellet cliff (Svalbard), Height is about 900m. The identification of the different layers and their shape on the textured model (right) , offer the possibility to calibrate and validate numerical model.

4.3. Risk mitigation infrastructures studies The threat of natural hazard on infrastructure such as road, railways, buildings, power lines requires some 3D data to run numerical simulation of the phenomenon so as to dimension correctly protection infrastructure. It is particularly important to get a high resolution 3D model and also its coverage to start with the most realistic terrain. For rock fall trajectory simulation (fig. 14a and 14b), LiDAR is used to model terrain with cliff and imagery permits, through textured mesh, to identify critical areas or unstable boulders to place simulation profile.

Figure 14a: 3D Model with overhang areas used for rockfall trajectory simulation, Tenay (FR). Oblique data acquisition. Point cloud filtering must be done manually since vegetation is present in the cliff. The cliff height is about 120m. The blue line represents a

Figure 14b: Oblique acquisition in Gorge de la Daille (FR) for rock fall protection. To the left, the ground point cloud represented in 2 colors (violet=overhang areas, orange=ground), in the center, the 3D mesh triangles, and to the right the textured mesh with

P4-1


simulation profile.

contour level overlay. The cliff height is about 250m. Acquisition time is 15 min flight.

4.4 Limitations The handheld helicopter base oblique data acquisition technique permits to cover all the complex terrain with a homogeneous accuracy, density and resolution. Nevertheless, the limits of the technique could be the following:

♦ Accuracy: with low height flight, mapping accuracy can reach ~5cm for each component. It is almost impossible for natural terrain to get better results than this threshold. Thus, the airborne technique is not relevant to detect very small motion or deformation (<10cm). Terrestrial survey (LiDAR, GPS…) is more suitable for such purpose.

♦ Weather: As fast as the deployment is concerned, airborne LiDAR/photogrammetry can not be used under rain, snow or cloud presence. The weather should be dry and free of clouds (between helicopter and object). The wind must be moderate so as to helicopter can follow smoothly the planned flight line. Flight altitude is not a real limitation since helicopters can fly until 5500 m above sea level.

♦ Surface: Handheld oblique data acquisition is relevant for small-medium surfaces. The limit depends on the complexity of terrain but above 5’000-8’000 ha, the need for 10cm range accuracy data must really be justified, otherwise, conventional vertical aircraft data acquisition is more relevant.

5. Conclusions LiDAR mapping technique is a very efficient tool for alpine environment mapping in such different domains as natural hazard study, geology or alpine environment monitoring. Thanks to handheld oblique data acquisition, accurate and high resolution topographic data over inaccessible areas and complex terrain can be measured. The high flexibility of the system allows mapping quickly any type of surface until the surface size of ~5000 ha. The LiDAR is also the best mapping tool to obtain terrain model under forested areas. Geosciences domains can benefit for such high resolution data for morphology analysis and phenomenon monitoring and studies. Nevertheless, end users must keep in mind that this technique has its accuracy limit in a range of 5-10cm. Deformation of few centimeters can not be monitored with airborne mapping. The huge amount of collected data shall draw the user attention. When we talk about LiDAR and imagery, we talk about several millions of points and several gigabytes of images. It is really important to define its need in terms of accuracy and of level of rendered detail, so as to define the point density and pixel size. Acting like “what can do the most can do the less” will add unnecessary data processing cost which is tied to amount of data.


References BUCKLEY S.J., VALLET J., WHEELER W. AND BRAATHEN A., 2008. Oblique helicopter-based Laser scanning for digital terrain modelling and visualisation of geological outcrops, The Internal Archives of the Photogrammetry, remote sensing and Spatial Information sciences, Beijing, Commission 4. GLENNIE, C.L., 2007. Rigorous 3D error analysis of kinematic scanning LIDAR systems. Journal of Applied Geodesy 1 (3), 147-157.

FAVEY E., 2001 Investigation and improvement of Airborne Laser Scanning technique for monitoring surface elevation changes of glaciers. Ph-D Thesis - Eidgeno ssische Technische Hochschule Zu rich ETHZ.

FISCHER L., EISENBEISS H., KÄÄB A., HUGGEL C., HAEBERLI W., 2011. Monitoring topographic changes in a periglacial high-mountain face using high-resolution DTMs, Monte Rosa East Face, Italian Alps, DOI: 10.1002/ppp.717

KNIEB U., BIEVRE G., JONGMANS J., PATHIER E., SCHWARTZ S., VILLEMIN T, 2009. Combined geophysical and remotesensing investigations to study the kinematics of two clayey landslides in the Trièves area (French Alps). EGU General Assembly 2009, Autriche.

KRAUS K., WALDHÄUSL P., 1996, Manual of photogrammetry, Edition Hermès, 1996

SCHENK, T., 2001. Modeling and analyzing systematic errors in airborne laser scanners. Technical Notes in Photogrammetry, vol 19. The Ohio State University, Columbus, USA, p. 46. SOVILLA, B.; BURLANDO, P.; BARTELT, P. 2006, Field experiments and numerical modeling of mass entrainment in snow avalanches. J. Geophys. Res. 111, F03007, doi:10.1029/2005JF000391: 16 p. 2006. SKALOUD J., VALLET J., KELLER K., VESSYERE G. AND KÖLBL O., 2005. Helimap System®: Rapid large scale mapping using handheld LiDAR/GPS/INS/CCD sensors on helicopters. ION GNSS 2005 Congress. Long Beach CA. 2005 VALLET J. 2002, Saisie de la couverture neigeuse de sites avalancheux par des techniques aéroportées. Thèse EPFL N° 2610. VALLET J., 2009. Cartographie Héliportée avec un système LiDAR-Photogrammétrique portable à la main: Expérience pratique en montagne, Analyse des performances et de la phase de calibration, Colloque technique laser pour l’étude des environnements naturels et urbains –SFPT, Le Mans - février 2009. VOSSELMAN, G., MAAS, H.-G. (Eds.), 2010. Airborne and Terrestrial Laser Scanning , Whittles Publising, ISBN 978-1904445-87-6, 320 p.

aerial oblique lidar-photogrammetry mapping for geology… · 2011-12-09 · 38th zlg eth (cas)...

Documents