Transcript Kyoto, Japan. 10 August 2010

Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
http://www.ual.es/GruposInv/ProyectoCostas/index.htm
SHADED-RELIEFS MATCHING AS AN EFFICIENT TECHNIQUE
FOR 3D GEO-REFERENCING OF HISTORICAL DIGITAL
ELEVATION MODELS
F.J. Aguilara, I. Fernándeza, M.A. Aguilara, J.L. Pérezb, J. Delgadob, J.G. Negreirosc
Dept. of Agricultural Engineering, Almería University, Spain
Dept. of Cartographic Engineering, Geodesy and Photogrammetry, Jaén University, Spain
C ISEGI – Nova de Lisboa University, Portugal
a
b
Corresponding Author: F.J. Aguilar ([email protected])
Kyoto, Japan. 10 August 2010
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
Nowadays Coastal Elevation Models production (e.g. for shoreline
extraction) is efficiently accomplished by means of LiDAR technology
which is contributing to a wide range of coastal scientific investigations
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
1
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
Because LiDAR is a relatively new technology, historical data beyond
the past decade are practically unavailable (LiDAR mapping systems
were not become available commercially till the late 90s).
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
1977 Agriculture Photogrammetric Flight
Approximated scale 1:18000
Analogic B&W flight
No camera calibration certificate
Focal length around 152,77 mm
Kyoto, Japan. 10 August 2010
2
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
ESTEREOMATCHING
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Z = f(x,y)
Coastal Elevation Model
Kyoto, Japan. 10 August 2010
3
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
The latter approach requires a number of ground control points (GCPs)
to compute the absolute orientation of every stereo pair, a surveying task
that usually becomes inefficient and costly because the difficulty to
accurately identify and survey a suitable set of ground points which could
be pointed on the corresponding historic photographs.
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
4
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
1:33000 scale
Kyoto, Japan. 10 August 2010
1:18000 scale
1:5000 scale
5
Digitized images
(photographs)
Research Project RNM 3575:
Multisource Geospatial Data Integration
Interior Orientation for each
and Mining for the Monitoring and
photograph
Modelling of Coastal Areas Evolution
and Vulnerability
Automatic Relative Orientation for
each stereo-pair
INTRODUCTION
Course Absolute Orientation through 2 approximated full points (X,Y,Z)
and 1 approximated Z point
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
Non-oriented DEM generation by
automatic image stereo matching
RESULTS &
DISCUSSION
Oriented
reference
DEM
CONCLUSIONS
Shaded-relief generation for a
predetermined solar azimuth and
solar elevation
To avoid the necessity of ground
control points, a new approach to
historical CEMs 3D geo-referencing
is proposed along this work.
Automatic Shaded-relief matching
to find image corresponding points
(pixels coordinates)
Georeferentiation: pixel coordinates
for matched points go to ground
geocentric coordinates
Iterative 7 parameters 3D Helmert
transformation between matched points
(ground coordinates) with residual
threshold control at each iteration
Transformation parameters
application to non-oriented DEM to
obtain final absolute orientation
Variation of solar
azimuth and elevation
Kyoto, Japan. 10 August 2010
Transformation to a common
geodetic reference system
(geocentric coordinates X, Y, Z))
Again?
6
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
BACK
Kyoto, Japan. 10 August 2010
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
Matched 3D points allow computing an iterative least squares
registration between both CEMs by means of a robust seven
parameters 3D Helmert transformation. The outliers found after
each iteration were discarded and not taken into account in the next
one by establishing a threshold value to avoid gross errors due to
landscape changes
CONCLUSIONS
X 
 a11
Y    a
 
 21
 Z 
a31
a12
a22
a32
a13   x  X 
a23   y   Y 
a33   z   Z 
BACK
Kyoto, Japan. 10 August 2010
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
VILLARICOS
NEW APPROACH
FUNDAMENTALS
Almanzora river mouth
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
Antas dry-ravine mouth
CONCLUSIONS
GARRUCHA
Kyoto, Japan. 10 August 2010
7
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
DSM 1977 (10x10 m grid spacing)
DTM 2001 (10X10 m grid spacing)
Photogrammetric Flight 1:18000 scale Photogrammetric Flight 1:20000 scale
produced by Junta de Andalucía©
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
8
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Pre-oriented model by means of automatic relative orientation.
Preliminary course-orientation.
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Average Error = 16.12 m
Maximum Error = 63.29 m
Minimum Error = -35.03 m
Standard deviation = 22. 15 m
Kyoto, Japan. 10 August 2010
9
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
135º solar azimuth and 45º solar elevation
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Automatic matching algorithm based
on the Scale Invariant Feature
Transform (SIFT*; Lowe, 2004) to
identify conjugated points in image
space (pixel coordinates).
26 conjugated points were correctly found
Kyoto, Japan. 10 August 2010
10
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Shaded-relief image matching. Results from 3D Helmert adjustment
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
Estimated parameters
Parameter
Solar azimuth 270º
Solar elevation 45º
Solar azimuth 135º
Solar elevation 45º
Value
Accuracy
Value
Accuracy
ΔX
-36.15 m
0.74 m
-38.81 m
0.99 m
ΔY
-8.35 m
0.77 m
-7.65 m
1.01 m
ΔZ
-10.97 m
0.74 m
-9.42 m
1.00 m
ΔΩ
0.0081º
0,00377º
0.0141º
0,00418º
ΔΦ
0.0174º
0,00172º
0.0162º
0,00372º
ΔΚ
-0.0116º
0,00489º
-0.0123º
0,00503º
λ
1.0006
0,00077
0.9954
0,00253
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
11
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Surface Matching Results between 1977 and 2001
(270º solar azimuth and 45º solar elevation shaded-relief)
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Average Error = -1.03 m
Maximum Error = 11.45 m
Minimum Error = -20.92 m
Standard deviation = 2.70 m
Kyoto, Japan. 10 August 2010
12
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Surface Matching Results between 1977 and 2001
(135º solar azimuth and 45º solar elevation shaded-relief)
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Average Error = -0.31 m
Maximum Error = 7.79 m
Minimum Error = -15.18 m
Standard deviation = 1.89 m
Kyoto, Japan. 10 August 2010
13
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Surface Matching Results between 1977 and 2001. Absolute vertical
residuals distribution (135º solar azimuth and 45º solar elevation shaded-relief)
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
14
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
Comparison between the 1977 Photogrammetrically Oriented DSM and the
Shaded-relief Matching Oriented DSM
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
DSM/DTM
comparison
Maximum
(m)
Minimum
(m)
Standard
deviation (m)
DSM (135º/45º) –
PhotoDSM (1977)
6.40
-6.05
1.57
DSM (240º/45º) –
PhotoDSM (1977)
5.28
-6.76
2.56
2001 DTM – 1977
PhotoDSM
8.34
-7.28
1.60
14
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
• The results obtained from this work may be deemed as very promising,
showing a good co-registration between reference and historical CEMs in
heavily developed coastal areas. The point is the high efficiency and
robustness demonstrated for historical CEM 3D geo-referencing when it was
compared to costly and time-consuming traditional methods such as
photogrammetric absolute orientation based on surveyed ground control points
and self-calibrating bundle adjustment techniques.
• As a further work, this preliminary approach could be used as a previous
course matching to be subsequently refined by 3D robust surface matching.
For instance our approach could be used as a first step headed up to later
apply a Least Z-Difference (LZD) based surface matching algorithm to refine
the initial matching as much as possible. This second step should include
weight functions based on M-estimators to make the computation more robust
and resisting to the presence of outliers
Kyoto, Japan. 10 August 2010
15
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
RESULTS &
DISCUSSION
CONCLUSIONS
Kyoto, Japan. 10 August 2010
Thank you very much for
your kind attention
16
Research Project RNM 3575:
Multisource Geospatial Data Integration
and Mining for the Monitoring and
Modelling of Coastal Areas Evolution
and Vulnerability
INTRODUCTION
NEW APPROACH
FUNDAMENTALS
STUDY SITE &
DATASETS
Pre-oriented model by
Shaded-relief image
matching
RESULTS &
DISCUSSION
CONCLUSIONS
Solution refining
Differential model
computation dZi
Tukey’s Biweight
(1  u 2 ) 2
w(u )i  
0
dz
con u  i
d i  dzi  dzvecindad ;
1 si d i  Med ( D) 
wi  

0 si d i  Med ( D)
M-estimator
Iterating till convergence
Binary weighting for
every point
Least Squares
estimation applying
weights (Helmert 3D)
Kyoto, Japan. 10 August 2010

 w dz
i
2
i
 min
para u  1

para u  1
Least squares
estimation applying
weights (Helmert 3D)
till convergence