Remotely sensed images provide up-to-date information for a wide range ... was not significant even the DEM generated from the stereo SPOT images are.
INVESTIGATING THE EFFECT OF DIGITAL ELEVATION MODEL ACCURACY ON THE PLANIMETRIC ACCURACY OF ORTHORECTIFIED SPOT IMAGERY
A THESIS SUBMITTED TO THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES OF THE MIDDLE EAST TECHNICAL UNIVERSITY
BY
MUSTAFA ERDOĞAN
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN THE DEPARTMENT OF GEODETIC AND GEOGRAPHIC INFORMATION TECHNOLOGIES
SEPTEMBER 2000
ii
ABSTRACT
INVESTIGATING THE EFFECT OF DIGITAL ELEVATION MODEL ACCURACY ON THE PLANIMETRIC ACCURACY OF ORTHORECTIFIED SPOT IMAGERY
ERDOĞAN, Mustafa M. Sc., Department of Geodetic and Geographic Information Technologies Supervisor: Assist. Prof. Dr. Mustafa TÜRKER
September 2000, 119 pages
Remotely sensed images provide up-to-date information for a wide range of applications. The recent emergence and widespread use of digital orthoimages are testimony to their importance. The digital orthoimages has become quite popular due to its diversity of use, particularly in their use as base information for Geographic Information Systems. In addition, the developments in image processing/analysis
techniques
for
automated
feature
extraction,
object
recognition and object classification make use of digital orthoimages.
One important factor that affects the planimetric accuracy of digital orthoimages is the quality of the digital elevation model (DEM). DEM generation constitutes a significant proportion of the total cost of producing digital orthoimages. Therefore, if DEMs generated from various sources can be used to
iii
orthorectify SPOT satellite images, then the orthoimage production cost can be reduced significantly.
This thesis investigated the effect of DEMs that are generated from various sources on the planimetric accuracy of SPOT orthoimages. Specifically, the effect of DEMs generated from stereo SPOT images, 1:25.000 scale topographic maps and 1:35.000 scale aerial photographs on the planimetric accuracy of SPOT orthoimages was investigated. The purpose of this investigation was to determine if there is a significant degradation in the planimetric accuracy of SPOT orthoimages when different quality DEMs are used in the orthorectification process. The study results indicate that the degradation in orthoimage accuracy was not significant even the DEM generated from the stereo SPOT images are used in the orthorectification process.
Keywords: SPOT satellite images, orthorectification, orthoimages, digital elevation models, accuracy assessment, empirical method
iv
ÖZ
SAYISAL ARAZİ MODELİ DOĞRULUĞUNUN SPOT ORTOFOTOLARININ PLANİMETRİK DOĞRULUĞU ÜZERİNDEKİ ETKİSİNİN ARAŞTIRILMASI
ERDOĞAN, Mustafa Yüksek Lisans, Jeodezi ve Coğrafi Bilgi Teknikleri Bölümü Tez Yöneticisi: Yrd. Doç. Dr. Mustafa TÜRKER
Eylül 2000, 119 sayfa
Uzaktan algılama ile elde edilen görüntüler çok geniş yelpazedeki uygulamalar için güncel bilgi ihtiyacını karşılarlar. Dijital ortofotoların mevcut hızlı ve yaygın kullanımı, bu ortofotoların öneminin göstergesidir. Dijital ortofotolar geniş kullanım alanları, özellikle Coğrafi Bilgi Sistemleri için altlık bilgi olarak kullanılmasıyla oldukça popüler olmuşlardır. Ayrıca, otomatik detay üretimi, detay tanıma ve detay sınıflandırması için görüntü işleme/analiz tekniklerindeki gelişmeler dijital ortofotolardan faydalanmışlardır.
Dijital ortofotoların planimetrik doğruluğunu etkileyen önemli bir faktör sayısal arazi modelinin (SAM) kalitesidir. SAM üretimi dijital ortofoto üretimi toplam maliyetinin önemli bir bölümünü oluşturmaktadır. Bu nedenle, farklı kaynaklardan üretilen SAMlar SPOT uydu görüntülerinin ortorektifikasyonu için kullanılabilirse, ortofoto üretim maliyeti önemli bir ölçüde düşürülebilecektir.
v
Bu tez çalışmasıyla, farklı kaynaklardan üretilen SAMların SPOT ortofotolarının planimetrik doğruluğu üzerindeki etkisi araştırılmıştır. Özellikle, stereo SPOT görüntüleri, 1:25.000 ölçekli topoğrafik haritalar ve 1:35.000 ölçekli hava fotoğraflarından üretilen SAMların SPOT ortofotolarının planimetrik doğruluğu üzerindeki etkisi araştırıldı. Bu araştırmanın amacı, ortorektifikasyon için farklı kalitedeki SAMlar kullanıldığı zaman SPOT ortofotolarının planimetrik doğruluğunda bir bozulma olup olmadığını belirlemektir. Çalışma sonuçları gösterdi ki, ortorektifikasyon için stereo SPOT görüntülerinden üretilen SAM kullanılsa bile ortofoto doğruluğunda dikkate değer bir azalma olmamaktadır.
Anahtar Kelimeler: SPOT uydu görüntüleri, ortorektifikasyon, ortofoto, sayısal arazi modeli, doğruluk araştırması, deneysel metot
vi
To my daughter and wife
vii
ACKNOWLEDGEMENTS
I would like to express sincere appreciation to my supervisor Assist. Prof. Dr. Mustafa TÜRKER for his guidance and valuable advice during the preparation of this thesis. Thanks go to Assoc. Prof. Dr. Nurünnisa Usul, Assoc. Prof. Dr. Vedat Toprak, Assoc. Prof. Dr. Nuri Merzi and Assoc. Prof. Dr. Volkan Atalay for their suggestions, comments and evaluation of thesis. I would like to express my sincere gratitude to the General Command of Mapping for offering its facilities during my study. I would like to submit my special appreciation to Dr. Mustafa ÖNDER and A. Selim TOPUZ for their help on my studies. Finally, I would like to thank my family for their encouragement and patience for my frequent absences throughout the research.
viii
TABLE OF CONTENTS
ABSTRACT ..............................................................................................
iii
ÖZ ..............................................................................................................
v
ACKNOWLEDGEMENTS ....................................................................
viii
TABLE OF CONTENTS ........................................................................
ix
LIST OF TABLES ...................................................................................
xii
LIST OF FIGURES .................................................................................
xiv
CHAPTER 1. INTRODUCTION .............................................................................
1
2. THEORY ............................................................................................
6
2.1 Introduction ....................................................................................
6
2.2 The Orthorectification Process ......................................................
6
2.3 Inputs to the Orthorectification Process ........................................
10
2.3.1 SPOT Satellite Data ..............................................................
10
2.3.2 The Role of Ground Control .................................................
19
2.3.3 The Role of DEM .................................................................
21
2.3.4 DEM Accuracy .....................................................................
22
2.4 Automatic DEM Generation ..........................................................
25
2.4.1 Image Matching Techniques .................................................
27
2.4.2 Problems Faced in Image Matching .....................................
31
2.5 Orthorectification Methods ............................................................
32
ix
3. METHODOLOGY ............................................................................
34
3.1 Introduction ....................................................................................
34
3.2 Study Overview .............................................................................
34
3.3 Study Area .....................................................................................
35
3.4 The Workflow .................................................................................
37
3.5 Data Inputs .....................................................................................
39
3.5.1 Image Data ............................................................................
39
3.5.2 Photogrammetric Process .....................................................
40
3.6 Data Processing .............................................................................
42
3.6.1 DEM Generation ...................................................................
42
3.6.1.1 DEM from 1:25.000 Scale Contour Lines ................
42
3.6.1.2 DEM from Aerial Photographs .................................
44
3.6.1.3 DEM from SPOT Images ..........................................
47
3.6.2 Orthophoto Generation .........................................................
55
3.6.2.1 Orthorectification of Aerial Photographs .................
55
3.6.2.2 Orthorectification of SPOT Images ..........................
57
4. RESULTS ...........................................................................................
64
4.1 Introduction ....................................................................................
64
4.2 Photogrammetric Process ..............................................................
64
4.3 Accuracy Assessment of DEMs ....................................................
66
4.4 Accuracy Assessment of Orthoimages ..........................................
76
5. CONCLUSIONS AND RECOMMENDATIONS ..........................
88
5.1 Summary ........................................................................................
88
5.2 Conclusions ....................................................................................
90
5.3 Recommendations ..........................................................................
92
x
REFERENCES ........................................................................................
93
APPENDIX A. Adjusted
Coordinates
and
Horizontal
Accuracies
of Aerial
Triangulation ........................................................................................
96
B. Exterior Orientation Parameters ............................................................
98
C. Adjusted Coordinates and Vertical Accuracies of Aerial Triangulation
99
D. DEM Measurements .............................................................................
100
E. DEM Error Profiles ...............................................................................
104
F. Point Feature Measurements on Orthoimages .......................................
113
G. Line Feature Measurements on Orthoimages .......................................
117
H. Area Feature Measurements on Orthoimages .......................................
119
xi
LIST OF TABLES
TABLE 2.1
Orbital characteristics of SPOT satellites .....................................
12
2.2
Radiometric characteristics of SPOT satellites ............................
13
2.3
HRV characteristics summary (SPOT-1,2,3) ...............................
15
3.1
SPOT imagery information ..........................................................
39
3.2
SPOT stereo model errors in DPW ...............................................
47
3.3
SPOT model errors in PCI before editing ....................................
50
3.4
SPOT model errors in PCI after editing .......................................
50
3.5
The results of automatic correlation of SPOT images in PCI ......
51
4.1
The height accuracies of 51 aerial triangulation points ................
67
4.2
Accuracies of DEMs .....................................................................
67
4.3
Accuracies of DEMs changing with slope ...................................
69
4.4
Accuracies of DEMs assessed by the comparison with reference DEM .............................................................................................
71
4.5
Errors for the profile in North-South direction .............................
75
4.6
Errors for the profile in East-West direction ................................
75
4.7
Errors for the profile along a road ................................................
76
4.8
The assessment results of orthoimages using point feature measurements ...............................................................................
4.9
The assessment results of orthoimages using line feature measurements ...............................................................................
4.10
79
The assessment results of orthoimages using area measurements of area features ...............................................................................
4.11
77
The assessment
results of orthoimages
using perimeter
measurements of the area features ..............................................
xii
81
82
A.1
Summary of the adjusted control point coordinates .....................
A.2
Adjusted coordinates and the horizontal accuracies of aerial
96
triangulation points .......................................................................
97
B.1
Exterior orientation parameters of aerial photographs .................
98
C.1
Adjusted coordinates and the vertical accuracies of aerial triangulation points .......................................................................
D.1
99
Measurements on DEM generated from 1:25.000 scale contour lines ...............................................................................................
100
D.2
Measurements on DEM generated from aerial photographs ........
101
D.3
Measurements on DEM generated from SPOT in PCI .................
102
D.4
Measurements on DEM generated from SPOT in DPW ..............
103
F.1
Point feature measurements on orthoimage generated using DEM from SPOT with automatic correlation in PCI .............................
F.2
Point feature measurements on orthoimage generated using DEM from 1:25.000 scale contour lines .................................................
F.3
114
Point feature measurements on orthoimage generated using DEM from aerial photographs with automatic correlation in DPW .......
F.4
113
115
Point feature measurements on orthoimage generated using DEM from SPOT with automatic correlation in DPW ..........................
116
G.1
Line Feature Measurements on Orthoimages ...............................
117
H.1
Area Feature Measurements on Orthoimages ..............................
119
H.2
Perimeter Measurements on Orthoimages ....................................
119
xiii
LIST OF FIGURES
FIGURE 2.1
An illustration of the concept of differential rectification .........
7
2.2
SPOT orbits ...............................................................................
13
2.3
Nadir viewing of SPOT .............................................................
16
2.4
Oblique viewing of SPOT .........................................................
16
2.5
Viewing geometry of SPOT ......................................................
17
2.6
Stereoscopic viewing capability of SPOT .................................
18
2.7
Principal of orthophoto generation ............................................
23
2.8
Effect of DEM errors on planimetric accuracy of the orthoimage .................................................................................
25
2.9
Example of an image pyramid ...................................................
27
2.10
Principal of cross correlation .....................................................
30
2.11
a. Left image; b. Right image; c. Cross correlation function .....
30
3.1
Study area ..................................................................................
36
3.2
A SPOT panchromatic image draped over an elevation model .
36
3.3
The workflow ............................................................................
38
3.4
An example regular aerial photo block ......................................
40
3.5
The DEM generated from 1:25.000 scale contour lines using
43
PCI EASI/PACE software ......................................................... 3.6
DEM production in Softplotter ..................................................
3.7
The DEM generated from aerial photographs in DPW using automatic correlation technique .................................................
3.8
45
46
The DEM generated from SPOT in DPW with automatic correlation technique .................................................................
48
3.9
DEM production in PCI .............................................................
49
3.10
SPOT images taken before and during highway construction ..
52
xiv
3.11
SPOT DEM with uncorrelated parts ..........................................
53
3.12
SPOT DEM after editing the uncorrelated parts .......................
54
3.13
Orthophoto production in DPW ................................................
56
3.14
A portion of orthorectified aerial photo .....................................
57
3.15
Orthoimage generated using DEM from 1:25.000 scale contour lines in PCI ................................................................................
3.16
Orthoimage generated using DEM from aerial photographs with automatic correlation in DPW ...........................................
3.17
61
Orthoimage generated using DEM from SPOT with automatic correlation in DPW ....................................................................
3.18
60
62
Orthoimage generated using DEM from SPOT with automatic correlation in PCI .......................................................................
63
4.1
Measurement points in aerial triangulation ..............................
65
4.2
The difference image between the reference DEM and the DEM generated from aerial photographs by automatic image correlation ..................................................................................
72
4.3
Error histogram of the difference image ....................................
72
4.4
The difference image between the reference DEM and the DEM generated from SPOT images in DPW by automatic image correlation .......................................................................
73
4.5
Error histogram of the difference image ....................................
73
4.6
The difference image between the reference DEM and the DEM generated from SPOT images in PCI by automatic image correlation ..................................................................................
74
4.7
Error histogram of the difference image ....................................
74
4.8
Point features used for accuracy assessment of the orthoimages
78
4.9
Line features used for accuracy assessment of the orthoimages
80
4.10
Area features used for accuracy assessment of the orthoimages
83
4.11
a. Normal orthoimage, b. Distorted orthoimage ........................
84
4.12
Merged contour lines over SPOT orthoimage ...........................
86
4.13
Merged line features over SPOT orthoimage ............................
87
xv
E.1
Error profile of DEM produced from aerial photographs with automatic correlation in DPW in North-South direction ...........
E.2
Error profile of DEM produced from SPOT in DPW in NorthSouth direction ...........................................................................
E.3
110
Error profile of DEM produced from SPOT in DPW along a road ............................................................................................
E.9
109
Error profile of DEM produced from aerial photographs with automatic correlation in DPW along a road ..............................
E.8
108
Error profile of DEM produced from SPOT in PCI in East-West direction .....................................................................................
E.7
107
Error profile of DEM produced from SPOT in DPW in EastWest direction ............................................................................
E.6
106
Error profile of DEM produced from aerial photographs with automatic correlation in DPW in East-West direction .............
E.5
105
Error profile of DEM produced from SPOT in PCI in NorthSouth direction ...........................................................................
E.4
104
111
Error profile of DEM produced from SPOT in PCI along a road ............................................................................................
xvi
112
Approval of the Graduate School of Natural and Applied Sciences
___________________ Prof. Dr. Tayfur Öztürk Director
I certify that this thesis satisfies all the requirements as a thesis for the degree of Master of Science
__________________________ Assoc. Prof. Dr. Nurünnisa Usul Head of Department
This is to certify that we have read this thesis and that in our opinion it is fully adequate, in scope and quality, as a thesis for the degree of Master of Science
__________________________ Assist. Prof. Dr. Mustafa Türker Supervisor
Examining Committee Members
Assoc. Prof. Dr. Nurünnisa Usul
__________________
Assoc. Prof. Dr. Vedat Toprak
__________________
Assoc. Prof. Dr. Nuri Merzi
__________________
Assoc. Prof. Dr. Volkan Atalay
__________________
Assist. Prof. Dr. Mustafa Türker
__________________
CHAPTER 1
INTRODUCTION
Remotely sensed images offer a unique perspective of the Earth, its resources, and human impact upon it. In little more than a decade, remote sensing technology has proven itself as a cost-effective source of valuable information for numerous applications including mapping, urban planning, environmental monitoring, agricultural management, oil exploration, market development, real estate siting and many others. In many ways, the value of remotely sensed satellite images has become obvious. They provide an overhead look at the features on Earth surface and help to understand relationships among these features. The practical value and applicability of remotely sensed data continue to grow rapidly as new and more advanced satellites are launched. With more satellites on the way, imagery will be available in an increasing selection of scene sizes, spectral resolutions, revisit frequencies, and spatial details. Up-to-date maps are the basic need in many applications. The need for upto date information drives the use of satellite imagery in mapping technology. Classical methods need years to produce a map, which is not preferable for the rapidly changing world. Remote sensing has become a solution for that problem and new technologies were developed through the up-to-date map needs. One of these technologies, which is very rapid and easy to use for mapping purposes, is the digital orthoimage, a photographic image that has been orthorectified to meet the precision and accuracy standards of a map.
1
In orthorectification, the distortions induced by the imaging platform, film and three-dimensional shape of the Earth are digitally removed from the space imagery. The final result is an image that has a precise geometry of a map which is a very popular product due to its diversity of use, particularly in its use as base information for Geographic Information Systems (GISs). Applications in many disciplines integrate the existing line data with digital images. One useful and very common integration is that of orthorectified images input directly into a GIS. This is quite advantageous for providing a base for a new data set or for updating existing databases. Another remarkable advantage of digital orthophotos is that they provide a more readily usable data source for a GIS when compared with conventional data sources. Because the information is not filtered through a cartographic interpretation, it remains unbiased. The relationships among land features such as buildings, transportation networks, etc. are presented in their natural form without being skewed by data conversions and interpretation. In addition, digital orthophotos are already in digital form and unlike hardcopy maps, they can be integrated directly into a GIS. Digital orthoimages serve as the backdrop with which older vector data can be updated or corrected. With the launch of new high-resolution satellites, an increasing number of orthoimages will be seen to be integrated into GISs. A significant portion of the total cost of producing digital orthoimages is that creating the Digital Elevation Model (DEM) required in the orthorectification process. Therefore, it seems logical to attempt to realize cost reductions from this part of the production workflow. If DEMs generated from various sources can be used to orthorectify medium spatial resolution image (5 to 100 m. spatial resolution) such as SPOT imagery, then the orthoimage production cost can be reduced significantly. Various studies were performed about DEM and orthoimage generation and their accuracy assessments. Different methods and data inputs were tested to achieve better accuracy. In one of these studies conducted by Chen and Lee (1993), an orthorectification method for SPOT images developed by the authors
2
was tested and the accuracies of the produced orthoimages were assessed. They found that accuracies better than two-thirds of a pixel could be achieved with their orthorectification methods. Another study conducted by El-Manadili and Novak (1996) was again the orthorectification process. They used a Direct Linear Transformation Model for precision rectification of SPOT Imagery. Especially, they worked on the effects of the number and quality of ground control points and base-to-height ratio of the images over the accuracy of produced orthoimage. The results show that sub-pixel accuracy can be achieved similar with the above study. Heipke et al. (1992) tested SPOT Imagery for point determination, DEM generation and orthorectification with the automatic photogrammetric processing. For the DEM they generated, an empirical height accuracy of better than 10 m. was achieved with small base-to-height ratios of 0.4 and 0.6. Giles and Franklin (1996) studied the evaluation of SPOT imagery for accuracy in elevation and its derivative topographic surfaces. Field measurements were used for the evaluation. They achieved a height accuracy of about 20 m. with a base-to-height ratio of 0.63 for DEM. Accuracies of DEMs derived from different data models were assessed by Li (1994). One of the DEMs was derived from photogrammetrically measured contour data and the other was the gridded data from aerial photographs. As a result, he suggested that contouring could be a better sampling strategy for better accuracy. The same kind of data was also used in this thesis. In a study conducted by Radhadevi et al. (1994), the geometric correction accuracy of SPOT stereo pairs was tested by using a single ground control point for orbit attitude modelling. Terrain coordinates are derived up to an accuracy of 28 meters in latitude and 40 meters in longitude and 27 meters in height with only 1 ground control point.
3
Day and Muller (1988) assessed the quality of DEMs produced by automatic stereo matchers from SPOT image pairs. DEMs were tested with manual photogrammetric measurements over aerial photography. They found RMSEs (root mean square error) of about 18, 10 and 80 meters with three different methods. A similar approach was carried in this thesis. Bolstad and Stowe (1994) produced DEMs from 1:40.000 scale aerial photographs and SPOT panchromatic pairs with automatic correlation technique and tested the accuracy of these DEMs and their derivative surfaces (slope, aspect). The produced DEMs were compared and differences of up to 82 meters were observed. Approximately 63 percent of the differences were 10 meters or less, and 90 percent were less than 22 meters. Pala and Pons (1995) tested the incorporation of relief in polynomial-based geometric correction of SPOT and Landsat TM imagery. It is investigated that sub-pixel accuracy can be reached with this method. Above studies show that the accuracy of DEMs, orthorectification process and the orthoimages are the concern of remote sensing society yet and many researches have been performed to develop the accuracies of these products, to reduce the production time and cost. Since SPOT imagery is widely used by public for DEM and orthoimage production, SPOT imagery has been used in many of these researches. The goal of this research is to study the effect of DEMs, which are generated from various sources, on the planimetric accuracy of digital SPOT orthoimages. Specifically, the effect of DEMs, which are generated from overlapping SPOT images, existing 1:25.000 scale topographic maps and 1:35.000 scale aerial photographs, on the planimetric accuracy of SPOT orthoimages was investigated. The purpose of this investigation is to determine if there is a significant degradation in the planimetric accuracy of SPOT orthoimages when the DEMs generated from various sources are used in the orthorectification process. If the DEM used in removing geometric errors in the image is accurate enough, then the orthoimages will offer outstanding consistency
4
over mountainous terrain. The accuracy of SPOT orthoimages that are to be generated will be assessed using an empirical method. The other objective of the study is to properly understand and appreciate the advantages of creating orthoimages from satellite images. This thesis is divided into five parts. In the following chapter (Chapter 2), the theoretical bases of the study are presented. Methods and inputs of orthorectification process and effects of them over this process are described. In particular, SPOT satellite system and stereoscopic aspects of this system are studied in detail. In Chapter 3, the methodology used in the study is presented. At the beginning, the study and the study area are represented. Data inputs and processes are defined. Since the methodology directly affects the accuracy results, it is very important to select an appropriate way to assess the planimetric accuracies of orthoimages. According to the needs and capabilities, the methodology was developed. In Chapter 4, measurements and accuracy results of photogrammetric process, DEMs and orthoimages are presented. Various measurements and various reference data were used for this purpose. Statistical accuracy assessments were made with the measurements and reference data. In Chapter 5, conclusions derived from the study and the recommendations that are thought to be useful guidelines for further studies on similar subjects are provided.
5
CHAPTER 2
THEORY
2.1 Introduction In this chapter, the theoretical basis of orthophoto production is presented. The chapter is organized into four sections. In the first section, the orthorectification process is outlined in general terms. In the second section, the main inputs to the digital orthorectification process are given. Specifically, SPOT satellite system is discussed in detail. The third section deals with automatic DEM generation from digital images. Finally, the last section includes the orthorectification methods available. 2.2 The Orthorectification Process As mentioned earlier, orthorectification is the process of removing relief displacement and sensor (camera) attitude variations from a space imagery (Jensen, 1996). The result is a planimetrically correct orthoimage on which direct measurements of terrain geographic location, distances, angles, and area can be made. On an unrectified imagery, such measurements can only be approximated due to relief and image displacements. The actual causes of variations in images differ for each type of sensor and therefore the procedure used to orthorectify space imagery depends directly on the type of sensor used to collect the image. The final product of the orthorectification is named as orthoimage or orthophoto.
6
The method illustrated in the Figure 2.1 is the process commonly referred to as differential rectification. In figure, point P is imaged at location P` on the mapping plane due to relief displacement. An orthogonal projection is achieved by placing a rectification plane at the average terrain height of the terrain falling within the bounds of the rectification segment containing point P.
Figure 2.1 An illustration of the concept of differential rectification.
In a digital context such as in softcopy photogrammetry, digital images are used in the process and are differentially transformed to a rectification plane on a pixel-by-pixel basis and digitally projected to a mapping plane thereby eliminating tilt and relief displacements. The terrain profile shown in Figure 2.1 can be considered to be a scan line in the digital image. If one-to-one correspondence between the orthophoto and the footprint of the raw image pixels is established, then one can imagine a rectification plane introduced into each
7
image pixel. The true orthographic positions of image points are determined by placing a rectification plane along a terrain profile at regular intervals called rectification segments. The rectification plane is placed at the average elevation of the terrain for each segment and the image data contained within the segment is rectified to this plane. Figure 2.1 illustrates this concept. The electronic projection of each pixel is accomplished using the wellknown mathematical models and the rectification plane is introduced at the appropriate terrain elevation with the aid of DEM. The image pixels are projected to the mapping plane, or equivalently the orthophoto matrix, through the assignment of a gray scale value to each grid element of a regular gridded DEM. This ensures that both the elevation and photographic density of ground surface elements are stored at the same planimetric location. If the gridded pattern of the DEM does not correspond to the pixel pattern of the raw image, then the elevations corresponding to each image pixel must be interpolated. The major advantage of orthophotos is that they can be produced in short terms in order to provide up-to-date information. This is especially true assuming that the terrain surface is subject to minor changes only and thus the DEM needed can be updated – where necessary – locally and used again. Up-to-date orthophotos can be applied e.g. for urgent planning purposes within short time, whereas elaborated line maps with a definitely longer production time would not be at hand. Dependent on the resolution, orthophotos provide more comprehensive and complete information than traditional vector maps. They are often used as stand-alone products leaving further interpretation and information extraction to the customer and his specialized staff.
However, if vector
information is necessary, it must be still extracted. Digital orthophotos offer the user the possibility to extract this information by himself. Besides the property of a shorter production time, orthophoto production is also less expensive than line mapping. This is due to its high degree of automation. If a customer cannot afford the time and/or the money for line map
8
production then the orthophoto data become of interest. The orthophoto data can be used as primary source of information either to focus the production of line maps or to focus their update if already existing. Furthermore, it provides complete georeferenced information for purposes of documentation. Orthophotos are also used in combination with vector data as backdrop. The geometric accuracy of orthophotos mainly depends on the quality of the underlying surface description. More accurate the topography is defined, then more accurate the orthophoto is. The definition of the topography can be made mainly with two data sources: (i) ground control, (ii) digital elevation model, which are the main inputs for the orthorectification process. Therefore, a user of orthophoto data should be aware of the effects inherent to orthophotos such as misplacement of objects that are not modelled by the DEM. Furthermore, the terms accuracy and pixel size should be strictly separated which are generally confused by the users. Metadata about the orthophoto including information about the input data directly connected to the orthophoto would be helpful in order to avoid the use of orthophotos for applications where the accuracy of orthophoto data is not sufficient. Orthophotos may be computed with a quite small pixel size for the sake of interpretation, but their positional accuracy may be less (Weidner, 1999). Orthoimages are the end product of orthorectification. Once created, these digital images can be merged with other data sources and mosaiced with adjacent orthoimages. The resulting digital file makes an ideal image backdrop for many applications,
including feature collection, visualization, and input
GIS/Remote Sensing systems.
9
into
2.3 Inputs to the Orthorectification Process As stated earlier, digital orthorectification process employs digital image processing operations. The images (images directly collected in digital form or film-based aerial images that are converted to digital form using high accuracy photogrammetric scanners) are the primary inputs to the digital orthorectification process. Other inputs include ground control points, measured image coordinates for the determination of the sensor parameters of each image, and a corresponding digital elevation model for the determination of the elevation of each pixel to be rectified in the image. Associated with each input is a degree of uncertainty that propagates errors into the final product. If care is taken to minimize the magnitude of uncertainty of the inputs, the resulting orthorectified image would constitute an accurate geometry. In the following subsections, these main inputs are explained in detail. First, the SPOT satellite system is reviewed. Next, some detailed description about the role of ground control points is provided. In the final subsection, the issues related to the use of a DEM in orthorectification process are revealed. 2.3.1 SPOT Satellite Data SPOT (Satellite Pour I’Observation de la Terre) satellite system has been developed by the French National Space Center (CNES) in collaboration with Belgium and Sweden. The first satellite in the program, SPOT-1, was launched on February 21, 1986, and has a spatial resolution of 10x10 m. (panchromatic mode) and 20x20 m. (multispectral mode). SPOT satellites 2 and 3 with identical payloads were launched in 1990 and 1993. In SPOT-4, which was launched in 1998, a number of design changes were made. These include the addition of a 20 meters resolution band in the mid-infrared portion of the spectrum, which was intended to improve vegetation monitoring, and mineral discrimination capabilities of data. Another change made for SPOT-4 was the addition of a separate wide field-of-view sensor called Vegetation Monitoring Instrument (WMI), which was primarily designed for vegetation monitoring.
10
SPOT satellites began a new era in space remote sensing, as it is the first satellite system to include a linear array sensor and employs pushbroom-scanning techniques. It is also the first system employing pointable optics. This enables side-to-side off-nadir viewing capabilities and it efforts full-scene stereoscopic imaging from two different tracks allowing coverage of the same area. SPOT-1 was decommissioned in December 1990 and remained in orbit on standby status for 7 years before being reactivated after the failure of SPOT-3. Planning for SPOT-5 is also underway. SPOT-5 will offer a greatly enhanced resolution while retaining the wide field of view. The multispectral resolution will be 10 meters while in panchromatic mode the resolution will be 5 meters. SPOT satellites are operated by the Centre National d'Etudes Spatiales (CNES), but the data are collected and distributed through a commercial network of private companies (SPOT IMAGE Corporation in U.S.A., SPOT IMAGE in France and SATIMAGE in Sweden). SPOT provides one of the most valuable sources of data in visible and near-infrared wavelengths for mainly three reasons that are;
it has a frequent repeat coverage, which increases the chances of capturing images in the visible wavelengths in areas with frequent cloud cover,
it provides stereoscopic viewing, which facilitates the creation of DEMs,
it offers one of the best spatial discrimination for urban studies of the medium spatial resolution (5-100 m.) satellite data available to the public.
The sensor payload for SPOT-1, -2, and –3 consists of two identical highresolution-visible (HRV) imaging systems and a package comprising two tape recorders and a telemeter transmitter. The satellite operates in a sun-synchronous,
11
near polar orbit (inclination of 98.7o) at an altitude of 832 km. The orbital characteristics and orbital routes of SPOT satellites are given in Table 2.1 and in Figure 2.2 respectively. The HRV sensors were designed to operate in two modes in the visible and reflective infrared portions of the spectrum: (i) a 10 m. resolution “panchromatic” (black and white) mode and (ii) a 20 m. resolution “multispectral” (color) mode. In panchromatic mode, imaging is performed in a single spectral band, corresponding to the visible part of the spectrum without the blue. The band covers the range 0.51 to 0.73 µm. This band is intended primarily for applications calling for fine geometrical detail. In multispectral mode, imaging is performed in three spectral bands, which are the Green band (0.50 to 0.59 µm), the Red band (0.61 to 0.68 µm) and the Near Infrared band (0.79 to 0.89 µm). The radiometric characteristics of SPOT satellites are given in Table 2.2.
Table 2.1 Orbital characteristics of SPOT satellites
Orbit shape
near circular
Equator crossing time
10:30 am (local time) descending node
Altitude
822 km
Inclination
98.7 degrees
Period (nominal)
101.4 minutes
Westward drift between ground tracks
2823 km
Repeat time
26 days
Orbits per cycle
369 (nominal)
Distance between reference tracks (at equator)
108.6 km
Data coverage
87 degrees North latitude to 87 degrees South latitude 12
Figure 2.2 SPOT orbits Table 2.2 Radiometric characteristics of SPOT satellites
Satellite
Band
Spectral Range Electromagnetic Region
XS1, Xi1
0.50 - 0.59 µm Visible Green
XS2, Xi2
0.61 - 0.68 µm Visible Red
XS3, Xi3
0.79 - 0.89 µm Near Infrared
SPOT 4
Xi4
1.58 - 1.75 µm Short Wave Infrared
SPOT 1 - 3
PAN
0.51 - 0.73 µm Visible
SPOT 4
MONO
0.61 - 0.68 µm Visible
SPOT 1 - 4
13
The HRV sensors of SPOT employ along-track or pushbroom scanning. It images the Earth with a linear array of charge-coupled-detectors (CCDs) arranged side by side along a line perpendicular to the satellite orbit track. A line of image data is obtained by sampling the response of the detectors along the array, and successive lines of data are obtained by repeated sampling along the array as the satellite moves over the Earth. Pushbroom scanning has three advantages over the mirror-sweep systems. First, fewer moving parts reduces the chances of sensor failure and increases the life of the sensor. Second, geometric errors that are commonly introduced by variations in mirror velocity when the mirror sweeps from one side to the other are avoided in array-detector systems. Finally, linear array detectors dwell on the area in their IFOV longer than sweeping systems do, thereby increasing the signal-to-noise ratio of the sensor (Lillesand and Kiefer, 1994). Each CCD array consists of 6000 detectors in panchromatic mode to record data at 10 m. resolution. Three 3000-element subarrays are employed in the multispectral mode at 20 m. resolution (Table 2.3). Data are encoded over a 256digital-number range. When looking directly at the Earth surface beneath the sensor system, the ground swath of each HRV scene is 60 km (Figure 2.3). In this configuration, the total swath width is 117 km. and the two scenes overlap by 3 km. However, with an oblique viewing capability of the sensors the mirror can be tilted to off-nadir viewing angles allowing each sensor to image a strip extending 475 km. to either side of the satellite ground track. In this configuration, it is possible to observe any region within a 950 km. wide strip centered on the satellite ground track (Figure 2.4 and 2.5).
14
Table 2.3 HRV characteristics summary (SPOT-1,2,3)
Multispectral mode
Panchromatic mode
Instrument Field-of-view
4.13 degrees
4.13 degrees
Pixel size
20 x 20 m
10 x 10 m
Number of pixels per line
3000
6000
Ground Swath Width (Vertical 60 km Viewing)
60 km
Ground Swath Width at maximum 80 km pointing angle (27 deg.)
80 km
Overlap between HRVs (vertical 3 km viewing)
3 km
Together, the two HRVs view a 117 km area with 3 km overlap.
15
Figure 2.3 Nadir viewing of SPOT
Figure 2.4 Oblique viewing of the SPOT
16
Figure 2.5 Viewing geometry of SPOT
The SPOT sensors can also acquire stereoscopic image pairs for a given geographic area. This is achieved by making two observations on successive days such that the two images are collected at angles on either side of the vertical (Figure 2.6). Stereoscopic imaging capability of SPOT allows generating DEMs from a pair of overlapping images. DEMs based on satellite images are essential for many applications when you need up-to-date and cost-effective information about terrain relief. Topographic mapping contouring and orthoimage generation are the two widely used application areas. A study conducted by Theodossiou and Dowman (1990) has shown that SPOT data could be used for mapping at 1:50.000 scale with 20-m. contours. And that if the data are very good and the ground control is sufficient, 1:25.000 scale plotting may be possible. Toutin and Beaudoin (1995) applied photogrammetric techniques to SPOT data and produced maps with planimetric accuracy of 12 m. with 90 percent confidence for well identifiable features and an elevation accuracy of 30 m. with 90 percent confidence.
17
Figure 2.6 Stereoscopic viewing capability of SPOT
Care must be taken however when selecting stereoscopic image pairs. The main factors that affect the quality of a DEM and must be taken into consideration are the base-to-height ratio (B/H), the difference in dates of image acquisitons, and the degree of overlap of the images. The ratio between the observation base B (distance between two satellite positions up to 850 km) and the height H (satellite elevation about 832 km) can be estimated using two values of the angle of incidence , supplied in the characteristics of each scene (GDTA, 1999). B/H=tanleft+tanright The angles of incidence of the two view are such that: left > 0 for the left view right< 0 for the right view
18
The ratio can be as high as 1.15 if angle of incidences are close to maximums. Such a high B/H ratio value can make digital automatic correlation difficult if the area contains high topographic relief since the number and size of concealed areas increase. The following B/H ratios may be chosen from SPOT programming order form:
1.0 (high value, applicable particularly for analog processing)
0.8 (average value, applicable for all application types)
0.6 (low value, useable in digital processing) The difference in dates between the acquisitions of two images must be as
short as possible in order to avoid excessive radiometric differences between two images. In addition, the seasons that the images are taken must be near, because images taken in different seasons become unlike and therefore it becomes very hard to process such kind of images. The size of the overlapping part of the two images depends particularly on the difference in orientation of the two images in the pair. It is evaluated:
either approximately by comparing the coordinates of the center of the scene on each image, and their geographic coverage and respective orientations,
or in more detail by comparing the geographic coordinates of the two scenes.
2.3.2 The Role of Ground Control The accuracy of orthoimages is a function of many variables, one of which is the accuracy of the ground control information used in a simultaneous adjustment and updating the satellite model parameters. The satellite model used in orthorectifying the images is a mathematical representation of the physical law
19
of the transformation between the image and ground spaces. It corrects the entire image globally and also takes into consideration the distortions due to terrain. Unlike the polynomial models, which require a large number of well distributed GCPs in order to avoid degradation of the model in some part of the image, the required number of GCPs is lower in the mathematical model. This is important when the control points are acquired by expensive differential GPS measurements. GCPs are used to calculate the position and orientation (i.e. roll, tilt, and yaw) of the imaging system at the moment of image taken. This calculation is accomplished using standard photogrammetric algorithms, such as a space resection or bundle adjustment. The position and orientation of the imaging system are expressed as six values: x, y, z, roll, tilt, and yaw (or alternatively, x, y, z, omega, phi, and kappa), which collectively define the exterior orientation of the imaging system for each image. They are needed in order to map each pixel of the digital image to its precise location on the ground. These processes can change according to the characteristics of the imaging system. The other issue to consider include the distribution of ground control in the image and the requirement for additional control points to provide redundancy. Ground control necessary for producing orthoimages often comes from ground surveying which would help to reduce the propagation of the error source into the orthophoto pixel positions. However, ground surveying techniques are usually costly. Alternatively, the ground control measurements may come from the aerial photographs or from hardcopy maps of the project area. As mentioned earlier, orthoimage generation uses the method of space resection to determine the relation between the object space and image space. Thus, the accuracy of the ground control used in the process effects the accuracy of the digital orthophoto. The absolute accuracy of an orthoimage depends also upon the quality of the ground control information.
20
From the foregoing, it is quite obvious that the role of ground control information in this study is twofold. First, ground control is required to relate the image space to the object space and second, independent checkpoints are required to ascertain the accuracy of the produced orthoimage. Any errors in the ground control information will be propagated into the pixel positions through the digital rectification process. It is therefore important to control this error source. 2.3.3 The Role of DEM A digital elevation model (DEM) is a digital representation of a portion of the Earth terrain over a two dimensional surface. The role of DEM in orthorectification process is that it eliminates terrain-induced displacements so as to transform an imagery into an orthogonal projection. The role of DEM is illustrated in Figure 2.7 where it can be seen that the DEM matrix, where individual elements have an elevation associated with them, can be made to correspond to a selected orthophoto matrix. Any ground sample distance (GSD) can be chosen for a given orthophoto. The GSD is then used to create an empty orthophoto matrix by basically dividing the specified study area into a regular gridded pattern. The digital imagery is orthorectified through the assignment of a gray-level value to each element in the DEM matrix. In fact, this is the essence of digital orthorectification whereby the pixels are projected to the specified orthophoto matrix through the assignment of a gray-level value to each grid element of the DEM. Thus, the role of DEM is twofold: (i) it sets the height of the elemental rectification planes (defined by the DEM grid interval) by storing the ground elevation corresponding to each plane and, (ii) it serves as the storage array for the gray-level values of each element. There are two methods for the assignment of gray-level values. These methods are discussed in section 2.5. DEM accuracy plays an important role in digital orthorectification process as it affects the accuracy of planimetry in the orthophoto. The accuracy of DEMs depends on a number of factors including source data scale, resolution and quality, DEM gridding interval and processing algorithms.
21
The accuracy of orthophoto is principally an inherent function of the accuracy of DEM, and subsequent interpolation errors. If there is one DEM point for every orthophoto pixel, no interpolation is required, and the interpolation errors are not introduced. Compiling this amount of DEM is unrealistic in a typical production environment and bigger intervals are used for DEMs (I.S.M., 1997). Having described the role of DEM in digital orthorectification process, it would be appropriate to discuss in more detail DEM accuracy and the effect of DEM inaccuracies on digital orthophotos. 2.3.4 DEM Accuracy The accuracy of a DEM is dependent upon its source and the spatial resolution, which is grid spacing, of the data profiles. The scale of the source data and resolution are also important factors that influence the accuracy of DEMs. A dependency exists between the scale of the source materials and the level of grid refinement possible. The source resolution is also a factor in determining the level of content that may be extracted during digitization. Another factor is the horizontal and vertical spacing of the DEM. Horizontal accuracy of DEM data is dependent upon the horizontal spacing of the elevation matrix. Within a standard DEM, most terrain features are generalized by being reduced to grid nodes spaced at regular intersections in the horizontal plane. This generalization reduces the ability to recover positions of specific features less than the internal spacing during testing and results in a de facto filtering or smoothing of the surface during gridding.
22
Figure 2.7 Principle of orthophoto generation (Wiesel, 1984)
23
Vertical accuracy of DEM data is dependent upon the spatial resolution (horizontal grid spacing), quality of the source data, collection and processing procedures, and digitizing systems. The entire DEM generation process, beginning with project authorization, compilation of the source data sets, and the final gridding process, must satisfy accuracy criteria needed for each application. Each source data set must qualify to be used in the next step of the process, since the errors have the effect of compounding for each step of the process. There are three types of vertical DEM errors; (i) blunder, (ii) systematic and (iii) random. Blunder errors are those errors of major proportions and are easily identified and removed during interactive editing. Systematic errors are those errors that follow some fixed pattern and are introduced by data collection procedures and systems. These error artifacts include: vertical elevation shifts, misinterpretation of terrain surface due to trees, buildings and shadows, and fictitious ridges, tops, benches or striations. Random errors result from unknown or accidental causes. These errors are reduced in magnitude by editing but cannot be completely eliminated (Idaho Geospatial Data Center, 1999). The RMSE is used to describe the vertical accuracy of a DEM, encompassing both random and systematic errors introduced during production of the data. Accuracy is computed by a comparison of linear interpolated elevations in the DEM with corresponding known elevations. Test points must be well distributed, representative of the terrain, and have true elevations with accuracies well known within the DEM accuracy criteria. Acceptable test points include, in order of preference: field control, aerotriangulated test points, spot elevations, or points on contours from existing source maps with appropriate contour interval. The error in the DEM will create a planimetric error on the orthoimage produced using this DEM. This error depends on different factors such that elevation error quantity, viewing geometry, imaging system etc. It can be basically calculated with the following formula and it is illustrated in Figure 2.8.
24
dx = dz*tan(i) where dx
=
planimetric error
dz
=
DEM error
i
=
viewing angle
i
True height Approximate height (DEM)
dz dx
Figure 2.8 Effect of DEM errors on planimetric accuracy of the orthoimage
2.4 Automatic DEM Generation DEMs can easily be calculated from stereo images using algorithms derived from automatic correlation methods. Automatic image correlation is based on simultaneous digital analysis of two overlapping images in a stereo pair. It is used for automatic production of DEMs. The process starts with modeling the sensor positions and attitudes for two views, then calculating the epipolar geometry. Next, one image of a pair is resampled in epipolar geometry. Correlation is performed among epipolar lines and for any reference point; the
25
best corresponding point is searched. The correlation results are then checked and corrected by smoothing, and if the results are inadequate, the process iterates to the previous step after redefinition of a better match. Last, the elevations are computed in mapping geometry. Measuring the similarity between the pixels is convenient (particularly in computing) to consider a square neighborhood. The window size is relatively empirical, and normally varies between 3X3 and 9X9. The analyst can choose the size as a function of experiences, the type and quality of the images, the type of relief (smaller windows for steep slopes), and the calculation time. To perform similarity measurement, one of the images is chosen as the reference image and a window is centered on the pixel for which the search is to be carried out. Then, the second image is explored by moving the search window and calculating the similarity index for each pixel position. The process continues until the pixel of the highest correlation is found. This however requires tremendous amount of calculations to be done. To speed the process, the images of the stereo pair are resampled to epipolar geometry and the search is performed on a single line. Furthermore, its size can be even further limited dynamically taking account of minimum and maximum elevations within the area. In addition, image pyramids can be used for calculating the similarity index faster. Commonly, the pixels in a neighborhood are highly correlated. In other words, most of the information is redundant and the search algorithms may take a lot of time. To reduce similarity index computation time, the search operation is performed on a pyramidal image. A pyramid is a generalized image structure consisting of several successively increased levels of resolution of one image (Figure 2.9). On a pyramid image structure so-called coarse-to-fine algorithms, which are vastly more efficient than their single level counterparts operating on a high resolution, are used. One main and important factor that affects the accuracy of automatically generated DEMs is the image matching technique employed.
26
Figure 2.9 Example of an image pyramid
2.4.1 Image Matching Techniques Image matching techniques are usually based on area based matching (ABM), and feature based matching (FBM). FBM uses complicated algorithms and is difficult to use. ABM is used more commonly and therefore it will be explained in detail. The idea of ABM is to shift and possibly warp one of the images such that its intensities best fit to the intensities of the other image. The best fit can be achieved using either a similarity or a distance measure. Small windows composed of pixels serve as matching primitives. The window center, possibly weighted, can be used for defining the location of a point to be matched. ABM has a high accuracy potential in well-textured image regions, and in some cases, the resulting accuracy can be quantified in terms of metric units. However, it is sensitive to the gray values that may change due to changes in illumination condition. In addition, the large search space for matching including various local extremes, the large data volume, blunders occurring in areas of occlusions, and 27
poor or repetitive texture are some of the difficulties that may be faced during area based matching. ABM is usually performed on local windows. Two approaches namely cross correlation and least squares matching are widely used. The cross correlation will be explained below since the software utilized in this study use cross correlation technique. The detailed description about the least squares matching is beyond the scope of this thesis. Technical detailed information about the least squares image matching can be found in Heipke, 96. In cross correlation, gray value models between two images depicting scene with a maximum correlation coefficient are found. The cross correlation coefficient is a simple but widely used measure for the similarity of different image windows. It is based on two assumptions: i.
The two images geometrically differ only due to translation, and
ii.
The two images radiometrically differ only due to brightness and contrast. Basic mathematical process steps of the cross correlation algorithm are
given below: p r u TG : q i c i v
; Where TG
= Geometric transformation between two images describing same
scene. u G are two unknown shifting parameters. v The radiometric transformation, T
Ti : h f a bf is linear with the parameters p I a, b .
28
The above equation shows correlation coefficient between the template window and the corresponding part of the search window
= Correlation coefficient
g1 r , c
= Individual gray values of template matrix
1
= Average gray value of template matrix
g 2 r , c
= Individual gray values of corresponding part of search
2
= Average gray value of corresponding part of search
R, C
= Number of rows and columns of template matrix
matrix
matrix
In order to compute the cross correlation function in two windows, a template window is shifted pixel by pixel across a larger search window (Figure 2.10), and in each position the cross correlation coefficient ρ between the template window and the search window is computed using equation-1. The maximum cross correlation coefficient defines the position of the best match between the template and the search window. A typical cross correlation operation is illustrated in figure 2.11 where a small template window of the left image of an aerial stereo pair and the corresponding larger search window in the right image are represented in Figure 2.11(a) and (b) respectively. In figure 2.11(c), a plot of the cross correlation function of the two windows is shown. As can be seen in figure 2.11(c), the spatial variation of the cross correlation coefficient can be extensive making it a difficult task to find its maximum.
29
(a) Template Window
(b) Search Window
Figure 2.10 Principle of cross correlation
(a)
(b)
(c)
Figure 2.11 a. Left image; b. Right image; c. Cross correlation function
30
2.4.2 Problems Faced in Image Matching Of a number of problems faced in image matching, similarity measure, regularization, algorithms complexity and approximate values are the important ones to be solved. The similarity measure of features within an image takes into account the uncertainty of features, the transformation and the type of the object. The definition of criteria for a good match obviously plays an important part in each matching algorithm. For Area Based Matching, the similarity between gray value windows is defined as a function of the differences between the corresponding gray values. This function can be the covariance or the cross correlation coefficient between the windows, the sum of the absolute differences between corresponding pixels, or - as is the case in least squares matching - the sum of the squares of the differences. Defining a similarity measure for feature based matching is more complicated. The definition must be based on the attributes of the features. In most Feature Based Matching algorithms, the differences in the geometric and radiometric attribute values are combined using heuristics and thresholds in order to compute the similarity measure, called a cost function or benefit function. Whereas a cost function is to be minimized, a benefit function must be maximized in order to achieve a good match. All matching problems are ill posed, that is they are underconstrainted, have no unique solution or have an instable solution. Therefore, they require regularization. For example, least squares estimation is a regularization method for in general underconstrained problems that have no unique solution. Algorithmic complexity problem comes from the very huge number of unknowns. In most matching and reconstruction problems, the number of unknowns and therefore the search space is huge. Generally, the number of unknown parameters is in the order of 104 – 107.
31
For faster calculations, approximate locations of the objects are needed. Using image pyramids are required to solve this problem. Approximate locations are calculated by using the reduced resolution images, and near this location, exact location is searched. 2.5 Orthorectification Methods The process of orthorectification differentially transforms, on a pixel-bypixel basis, a space imagery into an orthogonal projection. The correspondence between an image pixel and its corresponding ground element can be established in one of the two ways: (i) Direct orthorectification model: source image (l,p) geocoded image (i,j) (ii) Inverse orthorectification model: geocoded image (i,j) source image (l,p) In the direct method of orthorectification, the image coordinates of the center of the input pixel in the original digital image are used to calculate the output space coordinates of the corresponding ground element to which the image pixel gray value is transferred. This direct projection of each image pixel onto a mapping plane is accomplished using the reprojection case of collinearity equation. Although this direct method results in an output pixel for each input pixel, it creates a problem in that the elevation data required in the process must be stored in a rectangular matrix that is based on the digital image coordinates. This can be difficult to accomplish when a DEM has been collected from a source other than the image being orthorectified because the geographic extents of the image and the DEM may not correspond exactly and geocoding the image to the DEM can be problematic, as it must be performed in an off-line process. For these reasons, the indirect approach is the preferred method of orthorectification. The indirect or inverse model is the most convenient. It consists in computing, for each rectified pixel, the radiometric value according to the corresponding pixels of the raw image. Actually, to limit computing time without
32
noticeable errors, a regular grid can be used instead of applying the model for every image pixel. The grid is called Geometric Interpolation Grid (GIG). In that process, the grid-cell corner coordinates (I,J) are known. One pixel of the rectified image (i,j) is selected. Using the inverse model, the location of the each corner of the grid-cell is calculated (L,P) in the raw image. The location of the concerned pixel is calculated using a resampling method between the grid cell corners. The radiometric value of the rectified pixel is calculated using the adjoining pixels of the corresponding pixel in the raw image, by a resampling method.
33
CHAPTER 3
METHODOLOGY
3.1 Introduction In this chapter, the methodology used in this study is presented. First, a study overview is provided to give brief information about the methodology. Following this, the study area is summarized. Next, the workflow is given. Data inputs and processes used in the study and final products are provided in the following section. Finally, the DEMs and orthoimages produced, that are used in accuracy assessments, and their production steps are outlined. 3.2 Study Overview In this research, the effect of DEMs generated from various sources on the planimetric accuracy of orthoimages was evaluated by an empirical method in a study area near İzmir. The DEMs were produced from 1:25.000 scale contour lines, 1:35.000 scale aerial photographs by an automatic correlation technique in a digital photogrammetric workstation, and from a SPOT panchromatic stereo pair images using both a digital photogrammetric workstation and an image analysis system. After generating DEMs, the orthoimages were produced using these DEMs and their effects over the planimetric accuracy of orthoimages were analyzed.
34
3.3 Study Area The study area is situated in the West of İzmir, which is the third biggest city of Turkey (Figure 3.1). The area shows the characteristic of much of the Agean region, with agriculture and urban lands dominating valley bottoms and deciduous forests covering steeper areas. The area is generally rural. In the East of the study area there is a small village by which there exist some small farmlands. İzmir- Çeşme motorway and national road extend from West to Northeast of the study area. Besides the motorway, there are many roads all over the study area that can be detected in SPOT images and can be very useful for both ground control point selection and accuracy assessment. In the Northwest of the region, some small urban areas extend. The terrain varies between 0 and 750 meters in elevation. Majority of the area is open land with some single trees and brush. Slope changes gently. A SPOT panchromatic image draped over an elevation model of the study area is shown in Figure 3.2.
35
Figure 3.1 Study area
Figure 3.2 A SPOT panchromatic image draped over an elevation model
36
3.4 The Workflow The general workflow used in the study is shown in Figure 3.3. The workflow begins with the DEM production from different data sources using Vision Softplotter photogrammetry software and PCI EASI/PACE image analysis software. Software used have different DEM production algorithms and accuracy of DEM produced in that software is affected by these algorithms. Use of different software in DEM production for the same data source makes possible the comparison of automatic correlation algorithms of those software. Data sources used in the study are the stereo aerial photographs, SPOT images and 1:25.000 scale topographic map. The accuracy of DEMs, which are generated from these sources, would be expected to differentiate, since the source data of DEM is one of the most important factors for DEM accuracy. DEMs are generated by two methods namely manual digitizing and automatic correlation technique. The method used for DEM generation will also affect the accuracies of DEMs. Because of the above-summarized reasons, the accuracies of DEMs produced using different combinations of source data, method and software will also differentiate. Then, these DEMs were used for orthorectification of SPOT images and their effects on the planimetric accuracy of orthoimages are evaluated. In orthorectification process, the physical model computed using the ground control points and satellite ephemeris data is not changed, only DEM is changed, therefore, the source of the difference between the orthoimages will only be the effect of DEM used in orthorectification. Aerial triangulation is carried out and aerial photographs are orthorectified to use them as a reference data in the accuracy assessment. Aerial triangulation results and orthorectified aerial photographs are used for ground control point selection, which is necessary for both DEM productions from SPOT images and generation of the orthoimages, and accuracy assessment of DEMs and the orthoimages.
37
Aerial Photographs
Topographic Map (1:25.000)
Scanning
Digitizing
SPOT Images
DEM Generation
Digital Elevation Models
GCP Selection
Orthoimage Generation
Orthorectified SPOT Images Orthoimages
Figure 3.3 The workflow
38
Accuracy Assessment
3.5 Data Inputs Main data inputs for the study are image data and aerial triangulation. Image data are eight 1:35.000 scale black and white aerial photographs and two SPOT images. All of them are stereo images, thus can be used for DEM production. Aerial triangulation was carried out by using the field observations which were performed in 1996 for the revision of 1:25.000 scale maps of the area. 3.5.1 Image Data Aerial photographs were taken in 1996 by Zeiss RMK TOP15 camera (focal length: 153 mm) with 1:35 000 scale. They are black and white photographs and were scanned in 21 microns. SPOT images are dated as 11.09.1993 and 15.08.1991. The images are two years apart which is not a preferred condition. However, acquisition months are very close and there is no big seasonal difference between the images. The B/H ratio of the SPOT images is about 0.6 that is very appropriate for digital processing. Some characteristic information about the stereo SPOT data are given in Table 3.1.
Table 3.1 SPOT imagery information
Scene ID
S2H2930911093035
S2H2910815090732
Date
11/09/1993
15/08/1991
View Angle (degree) 25.63699654
-5,563
Center
Longitude 26.700556
26.821111
Center
Latitude 38.326667
38.326389
Level
1A
1A
39
3.5.2 Photogrammetric Process Aerial triangulation establishes the geometry of the camera relative to objects on the Earth’s surface. It is a very critical step of photogrammetric processing. By aerial triangulation, the exterior orientation parameters, the location and attitude (rotation angles) of the camera or sensor during the time of taking aerial photographs, and three-dimensional coordinates of unknown points are determined using ground control points and/or ephemeris information. It is an economic technique to measure a large amount of object points with very high accuracy. Aerial triangulation is normally carried out for a block of images, containing a minimum of two images (Figure 3.4). The strip, independent model, and bundle methods are the common approaches for implementing triangulation, of which bundle block adjustment is the most mathematically rigorous.
Figure 3.4 An example regular aerial photo block (Erdas Field Guide, 1997)
In this study, a reference data was needed to perform the accuracy test of DEMs and orthorectified SPOT images. As the reference data should be a few times more accurate then the products to be tested, the aerial photos were chosen as the reference data.
40
Eight aerial photographs covering the study area were used for aerial triangulation. The photogrammetric block was measured on Zeiss C115 Analytical Plotter. This plotter is capable of measuring ground control points (GCP) and tie points on stereo SPOT images and aerial photographs. First, the camera information was imported from the orient file. This data can be obtained from the camera calibration report. The control points on the block that will be measured were taken from the ground file. Aerial photographs were then put over the plates and the ID. of the photographs used was written. Next, the interior orientation, which defines the geometry within the camera, was performed using four fiducials. During triangulation, the interior orientation must be available in order to accurately define the external geometry of the camera. Distortion on the film and rectangularity must be less than 0.2 mm. and 0.001 radian after the measurements of the fiducials and the adjustment of the measurements. This was followed by the relative orientation that is performed to generate a stereo model. During relative orientation, y-parallaxes were corrected on six points over the image. Y-parallaxes on every point and mean errors were less than 8 and 5 microns after the adjustment. After that, the absolute orientation was carried out. At the first model, at least two planimetric and three height control points were measured. After that, tie points were measured. The results were calculated and controlled. For the second model, the left image was changed with the new one and the inner orientation for the new image was executed. The plotter mode was changed to pseudo and the other steps were repeated. At the end of measurement of every column, errors in the measurements were searched. Planimetric and height standard deviations were less than 15 microns and 25 microns in photograph scale. Photograph numbers of the block were 7953, 7954, 7955, 7956, 7997, 7998, 7999, 8000. Block consists of two columns with four photographs in each.
41
13, 19, 22, 15, 13, 20, 22, and 14 points were measured on the photos 7953, 7954, 7955, 7956, 7997, 7998, 7999, and 8000 respectively. Totally, 51 points were measured on the block. The ground coordinates of the control points and image coordinates of all the points are written to an ASCII file. This file was used for adjustment of the block with PAT/B software. 3.6 Data Processing 3.6.1 DEM Generation Four different DEMs were produced, two from the aerial photographs and two from the stereo SPOT images using a PCI EASI/PACE image analysis and Vision Softplotter digital photogrammetric software. 3.6.1.1 DEM From 1:25.000 Scale Contour Lines One important source for producing DEMs is 1:25.000 scale standard topographic maps. First, the contour line patterns of the topographic maps were scanned using the Context A0 scanner. The resulting raster file was displayed on the screen and the raster contour line image was converted into vector format automatically. The errors were edited on the screen. Each contour line was then assigned a height attribute. This was followed by generating a DEM using PCI EASI/PACE software. OrthoEngine module was used for this purpose. First, which vector levels will be used for DEM production was defined. Grid spacing and the projection of the final DEM were determined. Finally, a DEM was generated with 30-meter interval. The DEM produced is shown in Figure 3.5.
42
Figure 3.5 The DEM generated from 1:25.000 scale contour lines using PCI EASI/PACE software
43
3.6.1.2 DEM from Aerial Photographs The DEM generation process from aerial photos is based on an image matching technique mentioned earlier. In this study, Vision Softplotter digital photogrammetric workstation was used for this purpose. First, a project was created. Projection and some other parameters were defined before starting the processes. Eight aerial photographs were scanned in Zeiss Scai Film scanner in 21 microns. Data size of each scanned photograph was about 124 MB. DEM production workflow for Softplotter is shown in Figure 3.6. First, Block Tool module was used for import and interior orientation of the images, the triangulation process, and creation of support data for all images in the triangulation. When creating a new block the standard frame was chosen for the aerial photographs. A name was given to the block and reference frame was chosen. DEM Tool was used to automatically collect a user-defined ground space matrix of elevations from triangulated imagery. Interactive editing was performed with the stereoscopic display of the DEM over the imagery. The DEM points were checked by an operator to be exactly at the same height with the ground, not over or under the ground height. Because the accurate correlation could not be succeeded all over the area, some holes existed on the DEM. The holes detected were filled manually. Since the area is generally naked, the editing process did not take too much time. The DEM produced is shown in Figure 3.7.
44
Importing Imagery
Interior Orientation
Aerial Triangulation
Stereo Model Generation
DEM Generation
Figure 3.6 DEM production in Softplotter
45
Figure 3.7 The DEM generated from aerial photographs in DPW using automatic correlation technique
46
3.6.1.3 DEM from SPOT Images DEM from stereo SPOT images were produced using above named two systems. First, a DEM was generated using Vision Softplotter. The similar steps followed in DEM generation from aerial photographs were followed when generating the DEM from stereo SPOT images. SPOT images come with certain ephemeris data. Therefore, the user definition of the interior orientation was not required. However, for a better exterior orientation, the GCPs that are required were measured. Then, the ephemeris data and GCP measurements were triangulated and the block was accepted. Eleven full control points were chosen on both images. Two extra points were chosen on the left image out of the overlap area.
Totally, twenty-seven tie points were chosen for the stereo pair.
Triangulation was carried out in five iterations. The accuracies of the adjustment results are given in Table 3.2. After forming the model, a user-defined ground space matrix of elevations was automatically collected from stereo imagery using DEM tool. Because of the previously defined reasons, the DEM generated was then edited on stereo model. Holes detected by the operators were filled manually. The DEM generated is shown in Figure 3.8.
Table 3.2 SPOT stereo model errors in DPW
X(m)
Y(m)
S(m)
H(m)
Column Row (pixel)
(pixel)
Pixel
Mean Error
6.626
4.614
8.070
1.952
0.39
0.16
0.42
RMSE
8.201
5.144
9.840
2.107
0.49
0.35
0.60
47
Figure 3.8 The DEM generated from SPOT in DPW using automatic correlation technique
48
Upon DEM generation using Vision Softplotter, a DEM from SPOT images was generated next using PCI EASI/PACE software. The workflow is illustrated in Figure 3.9.
Image Import
GCP Collection
Model Calculation
Resampling in Epipolar Geometry
DEM Production
Figure 3.9 DEM production in PCI Ground control points were measured in GCPWorks module. From the setup menu “Add/Change GCP Only – Satellite Ortho Correction” choices were selected. Projection and datum information was determined. The projection and datum information used for the project area is: UTM 35S E004 (UTM: Universal Transversal Mercator Projection; 35: UTM Zone 35 with central meridian 27o East; S: Zone between 32o-40o North; E004: International 1909 ellipsoid). Well distributed 32 and 19 ground control points were measured on left and right images respectively. Then, a mathematical model that is necessary for orthorectification and DEM generation was formed in SMODEL module using satellite ephemeris data and ground control points. When required, ground control points were
49
interactively edited during the process. In editing, those points with errors bigger than 10 m. in any direction were either manually corrected or deleted. The processes were carried out for both images. The unedited and edited results are provided in Table 3.3 and Table 3.4 respectively.
Table 3.3 SPOT model errors in PCI before editing
Left Image (32 Points)
Right Image (19 Points)
X(m)
Y(m)
S(m)
X(m)
Y(m)
S(m)
8.05
8.26
11.53
3.98
3.10
5.05
RMSE
Table 3.4 SPOT model errors in PCI after editing
Left Image (32 Points)
Right Image (19 Points)
X(m)
Y(m)
S(m)
X(m)
Y(m)
S(m)
3.74
3.93
5.43
3.98
3.10
5.05
RMSE
As can be seen in Table 3.4, the errors are about half a pixel which is considered good for DEM production and orthoimage generation. Above accuracies do not change much if the number of GCPs is increased.
50
After that, left image was selected as the reference data and the right image was resampled to the epipolar geometry in SEPIPRO module. The reason for selecting the left image as the reference data was that it is more recent. In epipolar geometry, the search operation for image matching is conducted in X direction only eliminating the search in Y direction. This increases the speed and reliability of the process This was followed by generating a DEM from reference and epipolar images in SDEM module. Geometry of the images was determined by GCPs, ephemeris data of the image and shape information of the Earth. Matching process was made by automatic correlation and height of the matched points was computed. A 20-meter spacing DEM was generated using the above module. Various parameters were tried and the most appropriate set of parameters were accepted for the final product. The results are given in Table 3.5. As can be seen in the table, 95% of the pixels was matched by automatic correlation with an average score of 82 (maximum score:100) which is a quite successful result. Table 3.5 The results of automatic correlation of SPOT images in PCI
Average score for matching pixels
79.47
Average score for accepted pixels
82.49
Accepted pixels
95.47%
51
In certain parts, automatic correlation failed due to difference in spectral response in left and right images. For example, a highway construction that is missing in the early image appears in the late image (Figure 3.10). Along the highway therefore the images were not be able to correlated causing holes in the end product (Figure 3.11). Majority of those holes were manually edited by measuring their heights from 1:25.000 scale topographic map and replacing them in the corresponding places. Final product DEM is shown in Figure 3.12.
Figure 3.10 SPOT images taken before and during highway construction
52
Figure 3.11 SPOT DEM with uncorrelated parts
53
Figure 3.12 SPOT DEM after editing the uncorrelated parts
54
3.6.2 Orthophoto Generation After generating DEMs from different data sources, the orthoimages of SPOT data were generated next using those DEMs. As there are four DEMs, the orthoimages of SPOT data were produced four times using the same orbital model and their planimetric accuracies were checked. Any difference in planimetry between the SPOT orthoimages would be due to the effect of DEMs then. 3.6.2.1 Orthorectification of Aerial Photographs A reference data was needed to
accurately select
GCPs
for
orthorectification of SPOT images and to test their accuracies. Reference data may come from different sources such as; GPS observations or aerial photographs. In this study, the aerial photographs of the region were available and therefore they were used for the accuracy assessment since the spatial resolution and the accuracy of aerial orthoimages are much higher than that of SPOT orthoimages In order to make measurements on aerial photographs, they were orthorectified first in DPW using Ortho Tool module. The workflow is provided in Figure 3.13. Since a photogrammetric block was prepared earlier for DEM production from aerial photographs it was quite easy therefore to produce orthophotos in DPW. Thus, DPW was preferred for this production rather than PCI.
55
Importing Imagery
Aerial Triangulation
Stereo Model Generation
DEM Generation
Orthophoto Generation
Mosaicking
Figure 3.13 Orthophoto production in DPW
Totally, eight aerial photographs were orthorectified by using the aerial triangulation results and DEM from 1:25.000 scale contour lines which are produced from aerial photographs by manuel digitizing. The accuracy of orthophotos was expected to be about 1 meter. These orthophotos will then be used to assess the planimetric accuracy of SPOT orthoimages. A portion of the aerial orthophoto is shown in Figure 3.14.
56
Figure 3.14 A portion of orthorectified aerial photo 3.6.2.2 Orthorectification of SPOT Images After orthorectifying the aerial photographs to be used as reference data, the orthorectification of SPOT images was carried out in PCI. Orthorectification of satellite images is more complicated when compared with orthorectification of aerial photographs. This is due to the fact that geometric model of satellite images must correct the errors caused by satellite orbit and attitude variations, sensor geometry, Earth shape and rotation, and the relief. A commonly used modelling technique to correct the satellite images is to apply low order polynomials to the image. The advantage of this approach is that it involves a simple implementation. No knowledge about the satellite system is necessary. The polynomial is an improper model due to the errors caused by terrain displacements, imaging system characteristics, therefore a large number of well distributed ground control points are necessary to avoid degradation of the 57
model in some parts of the image. Orthoimages cannot be produced due to the high frequency nature of the terrain displacements which can not be modeled without DEM and imaging system information. A better result is achieved by combining a priori data with ground control point measurements in a simultaneous adjustment, and updating the satellite model parameters. Following this, the data only needs to be resampled once and the required number of ground control points will be lower. Since the viewing geometry is determined in the adjustment, a DEM can be used to eliminate terrain displacements, making orthoimage production possible. PCI uses the modelling method developed at the Canada Center for Remote Sensing (PCI, 1999). The major advantage of the method is that the satellite model is generic and can be applied to Landsat, SPOT, IRS and radar images such as RADARSAT, ERS-1 and J-ERS1. Moreover, models for other satellites and airbornes can also be added easily. The mathematical model was developed first to process raw SPOT images. It was then modified to process other visible satellite images, such as Landsat-TM and SAR images. The model is based on two co-linearity conditions, which represents the physical law of transformation between the image space and the ground space. It uses principles related to photogrammetry, orbitography, geodesy and cartography. The model reflects the physical reality of the complete viewing geometry and reflects all the distortions generated during the image formation as follows (PCI, 1999):
Distortions due to the platform (position, velocity and orientation),
Distortions due to the sensor (orientation, integration time and field of view),
Distortions due to the Earth (geoid, ellipsoid and relief)
Distortions due to the cartographic projection (ellipsoid and cartographic)
58
The modelling equations are simple and straightforward, with few unknowns requiring a few ground control points, as well as tie points if more than one image is used, to solve the equations. It has been proven that, the accuracy of the modelling is one-third of a pixel for passive sensor images and one pixel resolution for radar images, if the ground control points are good quality (PCI, 1999). As mentioned above, the model segment created by SMODEL module, which was used in DEM generation from stereo SPOT images, was with an accuracy about half a pixel. The previously produced four DEMs and satellite model segment created using the GCPs and ephemeris data were used for orthorectification. Orthorectification was performed by SORTHO module. There are a few parameters used in this module. These are input file, output file, DEM, model segment of the input file and resampling method. The two main inputs, which can affect the accuracy of orthoimage, are DEM and satellite model. Projection of the final orthoimage was determined by the GCP projection definition made in GCPWorks Module. Four different DEMs were used for the orthorectification of SPOT imagery. The only parameter changed in orthorectification process was the DEM. Modeling error was about half a pixel. Therefore, differences between the orthoimages would be only created by the differences between DEMs. The accuracy assessment of these orthoimages will help to present the effect of DEM errors over the orthorectified SPOT images. The produced final four orthoimages are shown in Figure 3.15, 3.16, 3.17 and 3.18.
59
Figure 3.15 Orthoimage generated using DEM from 1:25.000 scale contour lines in PCI
60
Figure 3.16 Orthoimage generated using DEM from aerial photographs with automatic correlation in DPW
61
Figure 3.17 Orthoimage generated using DEM from SPOT with automatic correlation in DPW
62
Figure 3.18 Orthoimage generated using DEM from SPOT with automatic correlation in PCI
63
CHAPTER 4
RESULTS
4.1 Introduction In this chapter, accuracy results calculated from the measurements on DEMs and orthoimages are outlined. The accuracies of aerial triangulation adjustment were presented in the first section. Next, measurements made on DEMs and the accuracies were presented. Finally, measurements made on SPOT orthoimages and aerial orthophotos and the accuracy assessment measurements were outlined. 4.2 Photogrammetric Process The residuals of the points measured in aerial triangulation were calculated with the adjustment. No point was detected with big errors. Horizontal and vertical accuracies are calculated with 24 points. Table A.1 in Appendix-A provides the summary of the adjusted coordinates of the control points used for the stereo model orientation. Figure 4.1 shows the distribution of all points used in six stereo models employed in the study.
64
Figure 4.1 Measurement points in aerial triangulation
65
The adjusted results for the control points were within the accuracy limits of 0.172 m. in planimetry and 0.204 m. in height. Maximum error in planimetry was 0.92 m. in one point. All the other errors were below 0.50 m. Minimum error was 0.07 m. In height, the maximum and minimum errors were 0.64 m. and 0.02 m. respectively. Table A.2 also provides the residuals of each control point in planimetry. The exterior orientation parameters, that are the coordinates of the image principal point and nine element of the rotation matrix for each aerial photograph, are given in Table B.1 in Appendix-B. 4.3 Accuracy Assessment of DEMs In total, four DEMs were produced. One was produced from 1:25.000 scale standard topographic maps. One was produced using the 1:35.000 scale stereo aerial photographs. Two were produced using the stereo SPOT images in two different software. All DEMs, except for the one from 1:25.000 scale standard topographic map, were generated using automatic image matching technique. To assess the accuracies of DEMs, three different approaches were used. In the first approach, DEMs were assessed using the 3D coordinates of 51 control points in aerial photo stereo models. The height accuracies of these points are given in Table 4.1 and their coordinates are listed in Table C.1 in Appendix-C. For the accuracy assessments, mean error and root mean square error (RMSE) were used. The formulas of these errors are given below: n
de Mean error
=
i 1
n n
de RMSE Where
=
2
i 1
n
de
:measurement error (measurement-reference)
n
:number of measurements
66
Table 4.1 The height accuracies of 51 aerial triangulation points
Mean Error
0.04 m.
RMSE
0.37 m.
As can be verified the errors are very small and therefore, these points were used as the reference data when assessing the DEM accuracies. The heights of the same points were read from the DEMs generated and they were compared with the heights of reference points. The summary of the results for the assessment is provided in Table 4.2. In table, the accuracies of the DEMs generated using SPOT imagery were summarized on the left columns. On the last two columns at the right side, the accuracies of the DEMs generated using the aerial photographs and 1:25.000 scale topographic map were summarized. All the measurements for the assessments are given in Appendix-D.
Table 4.2 Accuracies of DEMs
SPOT
DEM in DPW DEM from
PCI
DPW
aerial 1:25.000 scale
photo.
contour lines
Mean Error (m.) 7.07
4.94
3.28
2.87
RMSE (m.)
6.02
5.79
3.19
6.22
67
from
According to the results, the DEM generated from 1:25.000 scale contour lines has the highest accuracy. This is because the contour lines were drawn by the operators from 1:25.000 to 1:35.000 scale stereo aerial photographs using high precision stereoplotters. The DEM generated from aerial photographs with automatic correlation provided the second highest accuracy with 5.79 m. RMSE. In fact, these two DEMs have the same source. However, in the DEM produced with a fully automated way using automatic image matching techniques, the errors are slightly higher. Indeed, this was expected. The DEMs generated with automatic correlation from stereo SPOT images gave similar accuracies when compared with their RMSEs. DEM generated in DPW have better mean error however. No logical explanation can be made for this. Similar results were obtained in two previously made studies. In a study conducted by Heipke et al. (1992), a DEM generated from SPOT was checked. An empirical standard deviation of 10.8 meter was obtained with a small base-toheight ratio of 0.4. Since a better base-to-height ratio of 0.6 for the SPOT stereo pair was used in this thesis a RMSE of about 6 m. is a logical result. In another study conducted by Li (1994), DEMs generated from contour lines were searched. It was found that the expected accuracy of the DEMs (in terms of RMSE) derived from photogrammetrically measured contour data only was about CI (contour interval)/3 to CI/5 depending on the characteristics of the terrain topography. 1:25.000 scale contour lines are drawn with 10 m. interval. Therefore, the RMSE of DEM generated from this source should be between 2 m. and 3.3 m. according to the above study. RMSE of this DEM was computed as 3.19 m. which verifies the above study.
68
In a hilly topography, the slope may affect the accuracy of a DEM. Therefore, the study area was divided into user defined slope groups and the accuracy assessment was made according to these slope groups. In the region, the slope changes from 0 to 23. First, five groups were defined as 0-2, 3-5, 6-8, 9-11 and >11. It was then realized that there was not enough samples for some of the slope groups. Therefore, the groups were reduced to two only that are 0-7 and >7. The assessment results (Table 4.3) looked more meaningful with two slope groups.
Table 4.3 Accuracies of DEMs changing with slope
SPOT
DEM DPW
PCI Mean error(m.) RMSE (z) (m.)
DPW
in DEM
from
from 1:25.000 scale
aerial photo.
contour lines
0-7
6.18
5.28
2.93
2.45
>7
8.20
3.90
4.19
3.97
0-7
6.74
6.90
2.50
2.63
>7
5.14
4.14
6.10
4.32
It is known that if the slope gets steeper, the images taken from the different viewing directions would be seen different. This inevitably causes mismatching during automatic image correlation process. Even the visual interpretation may fail in such situations due to the difference in display. In other words, the steeper the slopes are, the higher the errors are expected. This trend can also be easily seen in Table 4.3. The RMSEs increase in the airborne DEMs with
69
the increase in slope. Surprisingly, RMSEs are nearly the same in spaceborne DEMs with the increase in slope. This may be due to the viewing geometry. As the airborne images are taken from very low altitudes with high B/H (base-toheight) ratios and big field of view angles, the slope becomes very effective on the display. Therefore, steeper slopes make the automatic correlation process more difficult causing accuracy to get lower. Spaceborne images are taken from very high altitudes with usually small B/H ratios and narrow field of view angles. For this reason, the images taken from different viewing angles do not become quite different from each other. Thus, the accuracies do not change much with the change in slope. In the second approach, DEMs were assessed by comparing them with a DEM having higher accuracy. According to the results of the first approach, the 1:25.000 scale DEM was selected as the reference data. To perform the accuracy assessment, the reference DEM points were subtracted from the corresponding DEM points generated with automatic correlation. For each DEM assessed, error statistics were calculated and given in Table 4.4. Difference images and their histograms are also shown in Figures 4.2, 4.3, 4.4, 4.5, 4.6 and 4.7 respectively. On the difference images, the white areas show the positive errors and black areas show the negative errors. The darker or the brighter the colors are, the higher the magnitude of errors is. Since these images are error images the histograms represent the error histograms.
70
Table 4.4 Accuracies of DEMs assessed by the comparison with reference DEM
SPOT
DEM in DPW from
PCI
DPW
photo.
Mean Error (m.)
7.63
7.93
3.96
RMSE (m.)
9.91
12.84
7.40
Max. Errors (m.)
-63/83
-95/134
-128/81
186216
325602
Number of points 186216
aerial
Since the number of points used in this assessment is very high the results are more logical and representative for the interpretation. As it was expected, DEM generated from aerial photographs have the smallest mean error. The mean errors from SPOT are quite close to each other. If the RMSEs are compared, again DEM generated from aerial photographs have the smallest RMSE. In addition, DEM generated from SPOT in PCI have a small RMSE and better maximum errors. But DEM generated from SPOT in DPW has higher RMSE and maximum errors. This was not an expected result. The only explanation may be done for this result is that PCI automatic image matching algorithm works better. Further researches are needed for a certain explanation, which is not the aim of this study.
71
Figure 4.2 The difference image between the reference DEM and the DEM generated from aerial photographs by automatic image correlation
Figure 4.3 Error histogram of the difference image 72
Figure 4.4 The difference image between the reference DEM and the DEM generated from SPOT images in DPW by automatic image correlation
Figure 4.5 Error histogram of the difference image 73
Figure 4.6 The difference image between the reference DEM and the DEM generated from SPOT images in PCI by automatic image correlation
Figure 4.7 Error histogram of the difference image
74
The third accuracy assessment was made with the profiles taken from the difference images of DEMs mentioned earlier. Two profiles were taken one in North-South direction and the other in East-West direction. The figures and the statistical values of the profiles are shown in Figures E.1, E.2, E.3, E.4, E.5 and E.6 in Appendix-E. Additional profiles were also taken along a main road. These profiles and their statistical values are shown in Figures E.7, E.8, and E.9 in Appendix-E. Minimum and maximum elevation errors, sample average error of each profile were calculated and given in Tables 4.5, 4.6 and 4.7 respectively.
Table 4.5 Errors for the profile in North-South direction
SPOT
DEM in DPW from
PCI
DPW
photo.
Average Error (m.) -3.61
0.85
1.72
Max. Errors (m.)
-22/38
-10/17
-24/30
aerial
Table 4.6 Errors for the profile in East-West direction
SPOT
DEM in DPW from
PCI
DPW
photo.
Average Error (m.) -5.55
-6.94
-4.08
Max. Errors (m.)
-29/14
-28/20
-17/6
75
aerial
Table 4.7 Errors for the profile along a road
SPOT
DEM in DPW from
PCI
DPW
photo.
Average Error (m.) 1.87
7.13
2.90
Max. Errors (m.)
-10/30
-8/15
-15/13
aerial
The results support the accuracies found in the previous approaches for the DEMs. The error profiles of DEMs generated with automatic correlation show some big variations. This is normal because mismatchings might have occurred during automatic correlation. If the profiles are compared, it can be easily seen that they show similar variations and the picks occurs in the similar locations. It means that software used in this study fail generally on the similar locations. 4.4 Accuracy Assessment of Orthoimages The final step in the workflow involved the assessment of the planimetric accuracy of the orthoimages generated. As there were four DEMs generated earlier, four orthoimages were produced in total therefore. As stated earlier, the aerial orthophotos were used as the reference data to perform the accuracy assessment. The assessment was carried out using the point, line and area features. First, the orthoimages were assessed using the point features. The coordinates of evenly distributed 45 sharp point features (Figure 4.8) were measured on each of the orthoimages and compared to their coordinates that were measured on the reference data set. The summary and the measurements of the assessment are provided in Table 4.8 and in Appendix-F respectively.
76
Table 4.8 The assessment results of orthoimages using point feature measurements
Errors
SPOT DEM in SPOT PCI
DEM Aerial
in DPW
Photo DEM
from
DEM in DPW 1:25K topo map
(m.) dy
ds
dx
dy
ds
from. dx dy
ds
contours dx dy
dx
Mean Error -
-
8.7
-
-
8.8
-
-
8.2
-
-
8.1
RMSE
6.8
6.1
7.9
7.4
6.5
6.9
7.2
6.2
6.6
7.5
6.0
dx
7.9
Surprisingly, the mean errors and RMSEs of the four orthoimages are nearly the same. The small differences between them may come from the point measurement precision. Therefore, the effect of the errors of DEMs over the orthoimages does not change the accuracy of the orthoimages. If the point results are compared, it is quite obvious that the point errors on all the orthoimages are similar. For example, the maximum error is about 30 m. at the first measurement point for all the orthoimages. All the DEMs affect the orthoimages in a similar way and there is no detectable difference between these DEMs for the role of them in orthorectification process. Above results closely match the results of a study conducted by Chen (1993) who generated orthoimages from the stereo SPOT data achieving 5.7 m. and 6.8 m. accuracy for rolling and rugged terrain respectively. The difference between the results of this study and his result may be due to the type of terrain and the quality of the images.
77
Figure 4.8 Point features used for accuracy assessment of the orthoimages
78
Second, the orthoimages were assessed using the line features. For the assessment, some linear features were digitized over the reference aerial orthophoto and SPOT orthoimages (Figure 4.9). 66 points were selected over the reference line with approximately 150 meters intervals. Every point was compared with the line features digitized from SPOT orthoimages. The shortest distances were measured between the points and line features of SPOT orthoimages whose planimetric accuracy would be assessed. The measurements and the test results are in Appendix-G and in Table 4.9.
Table 4.9 The assessment results of orthoimages using line features measurements
Errors
SPOT DEM SPOT DEM Aerial
(m)
in PCI
in DPW
Photo DEM
from
DEM in DPW
1:25K topo map
Mean Error
5.5
6.0
4.9
5.2
RMSE
4.3
4.5
4.1
3.5
The errors decreased in this test according to errors acquired using the point feature measurements. It is a logical and expected result, because the line segments intersects with each other and through the intersections, very small errors were acquired. This decreases the accuracies slightly. There is no big difference between the mean errors and RMSEs of orthoimages similar to the result of previous point feature measurements.
79
Figure 4.9 Line features used for accuracy assessment of the orthoimages
80
Finally, the orthoimages were assessed using the area features. First, common area features between the aerial orthophotos and SPOT orthoimages were searched. But, since these two data sources were two years apart from the each other, only eight area features with the clear borders could be detected (Figure 4.10). These features were drawn over orthoimages. While digitizing, the vectors drawn over other orthoimages were not used. Therefore, the vectors were made independent of each other. The areas and perimeters of the features were calculated and compared with ones over aerial orthophotos. The differences of the areas and perimeters and percentage of these differences were computed. The accuracy results are shown Table 4.10 and Table 4.11 in m2 and with percentage. The measurements are shown in Appendix-H.
Table 4.10 The assessment results of orthoimages using area measurements of the area features
Errors
SPOT DEM SPOT DEM Aerial in PCI
Mean Error (m2, %) RMSE (m2, %)
in DPW
Photo 1:25.000
DEM in DPW
contour line DEM
807.0
1355.1
1703.8
1674.9
%4.6
%7.8
%9.8
%9.6
824.3
1383.1
1498.3
1032.2
%4.7
%7.9
%8.6
%5.9
81
scale
Table 4.11 The assessment results of orthoimages using perimeter measurements of the area features
SPOT DEM SPOT DEM Aerial
Errors
in PCI Mean Error (m., %) RMSE (m., %)
in DPW
Photo 1:25.000
DEM in DPW
scale
contour line DEM
13.6
19.4
28.0
25.4
%2.5
%3.6
%5.2
%4.7
14.7
18.8
19.7
15.0
%2.7
%3.5
%3.6
%2.8
As can be seen, the orthoimages generated using the SPOT DEMs have slightly better results both in the area and perimeter measurements. Indeed, the contrary of this is an expected result. There is no logical explanation of these results. In the orthoimage that was produced by using DEM generated from stereo SPOT in PCI, a distortion was detected in one of the area (Figure 4.11). In such kind of productions, some local distortions may occur because of the use of a fully automated production method. This kind of errors can be corrected by the user interfaces in the needed steps and with editing after the production. However, when the time is a limited source, these automatic methods are always faster than the classical production methods. So, advantages and disadvantages of the methods should be thought before the production.
82
Figure 4.10 Area features used for accuracy assessment of the orthoimages
83
a
b
Figure 4.11 a. Normal orthoimage, b. Distorted orthoimage
84
All the above accuracy assessment approaches indicates that there is no identified important and clear difference between the orthoimages. It means that all the DEMs assessed in this study can be used for the orthorectification of SPOT images without loosing any in the accuracy. Assessment results also show that planimetric accuracies better than a pixel can be achieved using a well-defined topography with ground control points and digital elevation model. It means that SPOT system has an accurate physical modelling that can be used in orthorectification and makes possible such a better than pixel accuracy. Finally, 1:25.000 scale contour lines overlaid with orthoimages to see them superimposed (Figure 4.12). On this figure, contour lines have an proper imposition over the orthoimage and it can be easily seen over the valley bottoms. Some linear and area features were delineated from the aerial orthophotos. These features were also overlaid with the SPOT orthoimages (Figure 4.13). On this figure, it is quite obvious that line features have an accurate overlap over the orthoimage as it was expected from the previous planimetric accuracy assessment results. The most common use of orthoimages is the use of them in mapping applications. The acquired accuracies show that these orthoimages can be used for 1:50.000 or smaller scale topographical map production, which needs 10 m. or lower accuracies.
85
Figure 4.12 Merged contour lines over SPOT orthoimage
86
Figure 4.13 Merged line features over SPOT orthoimage
87
CHAPTER 5
CONCLUSIONS AND RECOMMENDATIONS
In the preceding chapters, the effect of DEMs generated from various sources on the planimetric accuracy of SPOT orthoimages was presented. In addition, general theoretical bases of automatic surface construction and orthoimage generation were provided. In this chapter, first the study and results were summarized and after that, the main conclusions that were reached during the course of this research are recapitulated, and recommendations concerning further studies are given. 5.1 SUMMARY In this study, the effect of DEMs generated from various data sources on the planimetric accuracy of SPOT orthoimages was investigated. The assessment of both DEMs and orthoimages was carried out using different approaches. First, DEMs which would be used for the orthorectification of SPOT image were generated. First DEM was generated from contour lines of 1:25.000 scale topographical maps. Second DEM was produced from 1:35.000 scale aerial photographs in DPW using automatic correlation techniques. The last two DEMs were generated from stereo SPOT images in DPW and PCI using automatic correlation techniques. After that the accuracy of DEMs were assessed. In the first approach of assessing the accuracy of DEMs, an accurate reference data set, the aerial photographs, were used.
The DEM generated from contour lines of
1:25.000 scale topo maps was found to be the most accurate one giving 3.19 m.
88
RMSE which is about one third of the 10 m. contour interval. The DEM generated from 1:35.000 scale stereo aerial photographs using automatic correlation technique resulted 5.79 m. RMSE. Considering that the DEM was generated by fully automated image matching techniques, this is an acceptable accuracy. Two DEMs were generated from stereo SPOT images by automatic image correlation using two different software that are the PCI EASI/PACE image analysis system and the Autometric Softplotter digital photogrammetric system. The RMSEs for both software were around 6 meters which are quite good for SPOT satellite images matching closely with the accuracies achieved by many other scientists. The effect of the slope over the accuracy of DEMs was also investigated. It was found that the accuracy of DEMs generated from aerial photographs decreased with the increase in slope. The reason for this is that since the aerial photographs were taken from low altitudes with small focal length aerial cameras, these overlapping areas with steep slopes do not become identical in the left and right images and therefore the automatic correlation fails easily in such areas. No significant difference was observed in the accuracy of DEMs generated from stereo SPOT images when analyzed with the change in slope. There is a logical explanation for this. Satellite images are taken from very high altitudes with lower spatial resolutions. Therefore, no significant difference occurs on steepy slopes in the left and right images that are taken from different viewing angles. In the second approach of assessing the DEMs, the DEM generated from 1:25.000 scale contour lines was used as a reference DEM and the others were subtracted from it. The assessment was then carried out on those difference DEM images. In this approach, the errors were computed over approximately 200.000 points. Similar results to the first approach were achieved. It was found that errors about 100 meters could occur in the DEMs produced with automatic correlation. In the third approach, profiles in North-South, East-West direction and along a road were taken from the difference images of DEMs mentioned above to visualize the errors. The accuracy results were verified by visual interpretation.
89
SPOT data was then orthorectified using above mentioned DEMs. As stated earlier, the aerial photographs were used as the reference data, and the planimetric accuracy assessment of the orthoimages were carried out using the point, line and area features. Basically, the orthoimages generated were compared with the reference data over a number of user defined sharp features to check their agreement in planimetry. The assessment using the point features was carried out over 45 well defined points. A RMSE of about 6 meters was found for all the orthoimages. This is about the half size of a SPOT pixel. There was no significant difference in the accuracies of orthoimages. This means that all the DEMs used in this study coming from various sources can be used for orthorectifying SPOT data without any decrease in the planimetric accuracy. The assessment using the line features was performed using the 66 points on the line features. A RMSE of about 4 meters was calculated for all the orthoimages. The orthoimage produced using the DEM that was generated from 1:25.000 scale contour lines gave slightly better results as this was the only DEM generated without fully automatic image matching techniques and believed to be more accurate than the others. The assessment using the area features was conducted over 8 area features since they were the only well identified area features in the study site. On the features tested, errors about 7% for the area of features and 3% for the perimeters of features were obtained. The number of area features was not enough for the accuracy assessment. Therefore, no logical explanation can be made over the results. 5.2 CONCLUSIONS In this study, some conclusions were derived from the results of the study. These are: It was concluded that the DEMs generated using manually digitized contour lines are more accurate than those generated by automatic correlation. However, there is a very high difference between the production times of these two techniques, which is a very important factor. 90
Automatic correlation can mismatch the pixels which is creating big errors while DEM generation and the automatic correlation algorithms need to be developed for higher accuracy and more homogenous DEMs. The produced DEMs should be controlled for holes and mismatchings and accurate DEM editing algorithms should be developed for these errors.
Some small distortions can occur over the orthoimages due to the local DEM errors. DEMs and the orthoimages generated using these DEMs should be controlled for such kind of errors.
The planimetric accuracies of produced orthoimages are better than a pixel. It means that SPOT imagery has a well-defined physical modelling that increases the accuracy of orthorectification process and needs lower number of ground control points.
The planimetric accuracy standard of 1:50.000 scale topographical maps is 10 m. Therefore, the accuracies of orthoimages obtained in this study comply with this standard and they can be used for the production of 1:50.000 scale topographical maps.
The final and most remarkable conclusion that can be made in this study is that the DEMs assessed in this study can be used in the orthorectification of SPOT satellite images without loosing much in the planimetric accuracy. On the other hand, the planimetric accuracy of around half a pixel can be achieved if above DEMs and accurate ground control points are used.
91
5.3 RECOMMENDATIONS The followings are recommended for further studies on a similar subject:
The study area used in this research is generally naked and rural. Only a few small parts are covered by the short brushwoods where, automatic image matching techniques easily fail. As well, the trees may also contribute errors in the height. Similarly, the buildings in the urban areas may affect the height accuracy. Therefore, the accuracy of DEMs in such areas would be lower and should be controlled more carefully.
The acquisition dates of right and left pairs of SPOT data that are used in DEM production should be close to each other. Otherwise, the changes on the features over the time would cause automatic correlation to fail and may create holes in the DEMs generated.
The seasons of the dates of images again should be near. Seasonal differences may create very different views in the images and therefore can make automatic correlation very difficult.
The
same
DEM
with
different
resolutions
can
be
used
for
orthorectification process and accuracy of the orthoimages can be assessed to define the appropriate resolution.
For a good physical modelling, accurate ground control points should be selected and these points should be well distributed over the image. The effect of the number of ground control points, accuracies of these points and their distribution over orthorectification process can be searched in a further study.
In this study, the accuracy assessment was performed over SPOT orthoimages. Stereo SPOT images can be used for accuracy tests and a comparison can be done between data collection over orthoimages and stereo models. 92
REFERENCES
1.
Bolstad P. V., Stowe T. (1994), An Evaluation of DEM Accuracy: Elevation, Slope and Aspect, Photogrammetric Engineering and Remote Sensing, Vol. 60, No. 11, pp 1327-1332
2.
Chen L., Lee L. (1993), Rigorous Generation of Digital Orthophotos from SPOT Images, Photogrammetric Engineering & Remote Sensing, Vol. 59, No. 5, May 1993, pp. 655-661
3.
Day T., Muller J. (1988), Quality Assessment Of Digital Elevation Models Produced By Automatic Stereomatchers From Spot Image Pairs, Photogramm. Rec., 12(72), pp 797-808
4.
Erdas Field Guide (1997), ERDAS Inc., Atlanta, USA
5.
GDTA (1999), Space Cartography Course Notes, May 1999, Toulouse, France
6.
Giles P. T., Franklin S. E. (1996), Comparison Of Derivative Topographic Surfaces Of A DEM Generated From Stereoscopic SPOT Images With Field Measurements, Photogrammetric Engineering And Remote Sensing, Vol. 62, No. 10, pp 1165-1171
7.
Heipke C. (1996), Overview of Image Matching Techniques, OEEPE, pp 173-189
93
8.
Heipke C., Kornus W., Strunz G., Thiemann R., Colomina I. (1992), Automatic Photogrammetric Processing of SPOT Imagery For Point Determination
and
Orthoprojecion,
International
Archives
of
Photogrammetry and Remote Sensing (29), B4, 465-471
9.
Idaho
Geospatial
Data
Center
(1999),
DEM
Processing
Steps,
http://geolibrary.mines.uidaho.edu/GeoLib/tutorial/dem_processing_steps.h tm, University of Idaho
10.
I.S.M. (International Systemap Corp.) (1997), The Fundamentals of Digital Photogrammetry, British Columbia, Canada
11.
Jensen J. R. (1996), Introductory Digital Image Processing - A Remote Sensing Perspective, Simon&Schuster/A Viacom Company, U.S.A
12.
Li Z. (1994), A Comparative Study Of The Accuracy Of Digital Terrain Models (DTMs) Based On Various Data Models, ISPRS Journal Of Photogrammetry And Remote Sensing, Vol. 49(1), pp 2-11.
13.
Lillesand and Kiefer (1994)
14.
Manadili Y., Novak K. (1996), Precision Rectification Of SPOT Imagery Using The Direct
Linear Transformation Model, Photogrammetric
Engineering And Remote Sensing, Vol. 62, No. 1, pp 67-72
15.
Pala V., Pons X. (1995), Incorporation of Relief in Polynomial -Based Geometric Corrections, Photogrammetric Engineering and Remote Sensing, Vol. 61, No. 7, pp 935-944
94
16.
PCI (1999), Manuel of Orthoengine Satellite Edition, PCI Inc., Ontario, Canada
17.
Radhadevi P. V., Sasikumar T. P., Ramachandran R. (1994), Orbit Attitude Modelling and Derivation of Ground Coordinates from SPOT Stereo Pairs, ISPRS Journal of Photogrammetry and Remote Sensing, 49(4), pp 22-28
18.
Theodossiou E. I., Dowman I. J.(1990), Heighting Accuracy of SPOT, Photogrammetric Engineering and Remote Sensing, Vol. 56, No. 11, pp1643-1649
19.
Toutin T., Beaudoin M. (1995), Real-Time Extraction of Planimetric and Altimetric Features from Digital Stereo SPOT Data Using a Digital Video Plotter, Photogrammetric Engineering and Remote Sensing, Vol. 61, No. 1, pp 63-68
20.
Weidner U. (1999), Practical Aspects of Digital Orthophoto Production, Hansa Luftbild GmbH, M nster, Germany
21.
Wiesel J.(1984), Digital Image Processing For Orthophoto Generation, Elsevier Science Publishers B.V., Amsterdam, Netherland
95
APPENDIX A
Table A.1 Summary of the adjusted control point coordinates
Planimetric
Height
Mean error (m)
0.291
0.307
RMSE (m)
0.172
0.204
96
Table A.2 Adjusted coordinates and the horizontal accuracies of aerial triangulation points
point-no. 79531 79532 79533 79541 79542 79543 79551 79552 79553 79561 79562 79563 79971 79972 79973 79981 79982 79983 79991 79992 79993 80001 80002 80003
x 457248.873 457162.245 456996.096 453751.882 453780.290 453752.692 450498.044 451032.518 450975.489 447677.033 447686.289 447042.340 449766.244 449205.827 448874.726 452389.941 452093.212 451679.327 455508.513 455369.043 455176.566 458700.593 459015.616 458960.845
y 4236824.965 4233375.820 4230205.248 4236812.314 4233169.986 4229944.467 4237069.730 4233596.440 4229819.026 4236421.850 4233582.548 4230743.389 4243916.201 4240643.591 4237553.739 4244381.122 4240643.343 4237022.573 4244103.743 4240782.844 4237279.479 4243994.649 4240648.704 4237100.190
rx 0.268 0.235 0.186 -0.120 -0.246 0.270 0.345 -0.183 -0.209 0.435 -0.212 -0.131 -0.019 -0.273 0.104 0.080 -0.112 0.124 0.073 -0.222 0.016 0.198 -0.081 -0.528 Mean error RMSE
97
ry 0.177 -0.047 -0.087 -0.023 -0.074 0.025 -0.858 0.134 0.018 -0.259 -0.051 0.293 -0.242 0.140 0.159 0.354 0.136 0.028 0.020 0.287 0.246 -0.121 -0.291 0.037
rs 0.321 0.240 0.205 0.122 0.257 0.271 0.925 0.227 0.210 0.506 0.218 0.321 0.243 0.307 0.190 0.363 0.176 0.127 0.076 0.363 0.247 0.232 0.302 0.529 0.291 0.172
APPENDIX B Table B.1 Exterior orientation parameters of aerial photographs photo-no. 7997
7998
7955
7956
7999
7954
7953
8000
omega -0.16 -0.996834 0.079103 -0.008001 0.29 -0.996944 0.077992 0.004394 0.32 -0.999120 0.041428 0.006561 -0.61 -0.999186 0.040303 0.001691 -0.65 -0.999183 0.040298 0.003073 0.17 -0.999006 0.044564 0.000536 0.27 -0.999947 0.009933 0.002770 -0.65 -0.999906 0.011382 0.007701
phi kappa rotation matrix -0.51 -194.96 -0.079125 -0.007778 -0.996862 0.003122 0.002496 0.999965 0.28 -195.03 -0.078011 0.004028 -0.996942 -0.004857 -0.004528 0.999980 0.42 -197.36 -0.041461 0.006347 -0.999128 -0.005284 -0.005016 0.999966 0.11 -197.43 -0.040285 0.002078 -0.999142 0.009573 0.009648 0.999952 0.20 -197.43 -0.040265 0.003480 -0.999137 0.010019 0.010151 0.999944 0.03 -197.16 -0.044565 0.000419 -0.999003 -0.002649 -0.002627 0.999996 0.18 -199.37 -0.009945 0.002728 -0.999942 -0.004269 -0.004242 0.999987 0.49 -199.28 -0.011303 0.007815 -0.999884 0.010074 0.010162 0.999919
98
px,py,pz 448743.346 4240921.576 5794.739 452034.268 4240733.955 5811.246 450420.981 4233466.202 5800.204 447125.010 4233365.066 5802.625 455363.799 4240598.335 5810.808 453702.708 4233551.853 5803.514 457010.168 4233611.620 5802.908 458729.164 4240569.899 5809.805
APPENDIX C
Table C.1 Adjusted coordinates and the vertical accuracies of aerial triangulation points Point-no. 79531 79532 79533 79541 79542 79543 79551 79552 79553 79561 79562 79563 79971 79972 79973 79981 79982 79983 79991 79992 79993 80001 80002 80003
X 457248.873 457162.245 456996.096 453751.882 453780.290 453752.692 450498.044 451032.518 450975.489 447677.033 447686.289 447042.340 449766.244 449205.827 448874.726 452389.941 452093.212 451679.327 455508.513 455369.043 455176.566 458700.593 459015.616 458960.845
Y 4236824.965 4233375.820 4230205.248 4236812.314 4233169.986 4229944.467 4237069.730 4233596.440 4229819.026 4236421.850 4233582.548 4230743.389 4243916.201 4240643.591 4237553.739 4244381.122 4240643.343 4237022.573 4244103.743 4240782.844 4237279.479 4243994.649 4240648.704 4237100.190
99
z 297.866 69.651 213.421 200.879 47.608 36.865 67.351 35.523 3.130 32.280 3.922 0.104 3.186 49.314 10.760 4.777 192.565 118.044 81.211 220.329 289.028 173.146 366.994 401.704 Mean error RMSE
rz -0.020 0.513 -0.184 -0.058 0.341 -0.102 0.619 0.645 -0.156 -0.447 0.403 -0.368 -0.295 -0.021 -0.274 0.075 -0.321 -0.096 0.541 0.549 -0.303 -0.069 -0.361 -0.610 0.307 0.204
APPENDIX D Table D.1 Measurements on DEM generated from 1:25.000 scale contour lines point no 320 328 863 79531 79532 79533 79541 79542 79543 79551 79552 79553 79561 79562 79563 79971 79972 79973 79981 79982 79983 79991 79992 79993 80001 80002 80003 100077 100078 100079 100080 100081 100082 100083 100084 100089 100090 100091 100092 100093 100100 100101 100102 100103 100104 100105 100106 100107 300031 300033 300034
x 448229.4 448367.5 457660.9 457248.9 457162.2 456996.1 453751.9 453780.3 453752.7 450498 451032.5 450975.5 447677 447686.3 447042.3 449766.2 449205.8 448874.7 452389.9 452093.2 451679.3 455508.5 455369 455176.6 458700.6 459015.6 458960.8 449766.2 449104.2 452390 452574.6 455508.5 455798.3 458700.6 459159.3 451679.5 453751.7 455176.5 457248.9 458960.8 446616.4 447042.3 450062.3 450975.5 453248.6 453752.7 456631.3 456996.2 446739.6 448982.4 448006.6
y 4235270 4230512 4237200 4236825 4233376 4230205 4236812 4233170 4229944 4237070 4233596 4229819 4236422 4233583 4230743 4243916 4240644 4237554 4244381 4240643 4237023 4244104 4240783 4237279 4243995 4240649 4237100 4243916 4242513 4244381 4242660 4244104 4242871 4243995 4242646 4237023 4236812 4237280 4236825 4237100 4231054 4230743 4230972 4229819 4230911 4229944 4231286 4230205 4236621 4237188 4236895
z 215.015 95.945 394.575 297.866 69.651 213.421 200.879 47.608 36.865 67.351 35.523 3.13 32.28 3.922 0.104 3.186 49.314 10.76 4.777 192.565 118.044 81.211 220.329 289.028 173.146 366.994 401.704 3.167 6.782 4.664 99.425 81.333 109.029 173.114 322.849 118.124 200.981 289.402 298.244 401.592 18.379 -0.061 12.593 3.106 55.988 36.846 54.338 213.048 -1.73 37.647 37.45 mean error RMSE
100
z-dem 216 93 390 299 70 217 199 50 41 71 40 4 37 8 1 2 53 13 8 190 120 83 224 289 183 371 402 2 12 6 100 83 110 183 328 120 199 289 299 402 13 1 15 4 60 41 50 217 5 41 34
dz -0.985 2.945 4.575 -1.134 -0.349 -3.579 1.879 -2.392 -4.135 -3.649 -4.477 -0.87 -4.72 -4.078 -0.896 1.186 -3.686 -2.24 -3.223 2.565 -1.956 -1.789 -3.671 0.028 -9.854 -4.006 -0.296 1.167 -5.218 -1.336 -0.575 -1.667 -0.971 -9.886 -5.151 -1.876 1.981 0.402 -0.756 -0.408 5.379 -1.061 -2.407 -0.894 -4.012 -4.154 4.338 -3.952 -6.73 -3.353 3.45 2.868 3.190
Table D.2 Measurements on DEM generated from aerial photographs point no 320 328 863 79531 79532 79533 79541 79542 79543 79551 79552 79553 79561 79562 79563 79971 79972 79973 79981 79982 79983 79991 79992 79993 80001 80002 80003 100077 100078 100079 100080 100081 100082 100083 100084 100089 100090 100091 100092 100093 100100 100101 100102 100103 100104 100105 100106 100107 300031 300033 300034
x 448229.4 448367.5 457660.9 457248.9 457162.2 456996.1 453751.9 453780.3 453752.7 450498 451032.5 450975.5 447677 447686.3 447042.3 449766.2 449205.8 448874.7 452389.9 452093.2 451679.3 455508.5 455369 455176.6 458700.6 459015.6 458960.8 449766.2 449104.2 452390 452574.6 455508.5 455798.3 458700.6 459159.3 451679.5 453751.7 455176.5 457248.9 458960.8 446616.4 447042.3 450062.3 450975.5 453248.6 453752.7 456631.3 456996.2 446739.6 448982.4 448006.6
y 4235270 4230512 4237200 4236825 4233376 4230205 4236812 4233170 4229944 4237070 4233596 4229819 4236422 4233583 4230743 4243916 4240644 4237554 4244381 4240643 4237023 4244104 4240783 4237279 4243995 4240649 4237100 4243916 4242513 4244381 4242660 4244104 4242871 4243995 4242646 4237023 4236812 4237280 4236825 4237100 4231054 4230743 4230972 4229819 4230911 4229944 4231286 4230205 4236621 4237188 4236895
z 215.015 95.945 394.575 297.866 69.651 213.421 200.879 47.608 36.865 67.351 35.523 3.13 32.28 3.922 0.104 3.186 49.314 10.76 4.777 192.565 118.044 81.211 220.329 289.028 173.146 366.994 401.704 3.167 6.782 4.664 99.425 81.333 109.029 173.114 322.849 118.124 200.981 289.402 298.244 401.592 18.379 -0.061 12.593 3.106 55.988 36.846 54.338 213.048 -1.73 37.647 37.45 mean error RMSE
101
z-dem 212 93 393 297 73 212 203 47 44 66 37 5 28 8 -1 4 50 12 7 190 113 80 224 287 181 367 399 4 10 7 100 80 111 181 325 113 203 287 297 399 17 -1 15 5 58 44 56 212 30 37 22
dz 3.015 2.945 1.575 0.866 -3.349 1.421 -2.121 0.608 -7.135 1.351 -1.477 -1.87 4.28 -4.078 1.104 -0.814 -0.686 -1.24 -2.223 2.565 5.044 1.211 -3.671 2.028 -7.854 -0.006 2.704 -0.833 -3.218 -2.336 -0.575 1.333 -1.971 -7.886 -2.151 5.124 -2.019 2.402 1.244 2.592 1.379 0.939 -2.407 -1.894 -2.012 -7.154 -1.662 1.048 -31.73 0.647 15.45 3.279 5.794
Table D.3 Measurements on DEM generated from SPOT in PCI
point no 863 79531 79532 79533 79541 79542 79543 79991 79992 79993 80001 80002 80003 100081 100082 100083 100084 100090 100091 100092 100093 100104 100105 100106 100107
x 457660.9 457248.9 457162.2 456996.1 453751.9 453780.3 453752.7 455508.5 455369 455176.6 458700.6 459015.6 458960.8 455508.5 455798.3 458700.6 459159.3 453751.7 455176.5 457248.9 458960.8 453248.6 453752.7 456631.3 456996.2
y 4237200 4236825 4233376 4230205 4236812 4233170 4229944 4244104 4240783 4237279 4243995 4240649 4237100 4244104 4242871 4243995 4242646 4236812 4237280 4236825 4237100 4230911 4229944 4231286 4230205
z z-dem 394.575 402 297.866 303 69.651 75 213.421 230 200.879 201 47.608 62 36.865 45 81.211 83 220.329 211 289.028 300 173.146 176 366.994 373 401.704 412 81.333 83 109.029 108 173.114 176 322.849 326 200.981 201 289.402 300 298.244 303 401.592 412 55.988 72 36.846 45 54.338 57 213.048 230 Mean error RMSE
102
dz -7.425 -5.134 -5.349 -16.579 -0.121 -14.392 -8.135 -1.789 9.329 -10.972 -2.854 -6.006 -10.296 -1.667 1.029 -2.886 -3.151 -0.019 -10.598 -4.756 -10.408 -16.012 -8.154 -2.662 -16.952 7.067 6.221
Table D.4 Measurements on DEM generated from SPOT in DPW point no 79531 79532 79533 79541 79542 79543 79552 79553 79983 79991 79992 79993 80001 80002 80003 100081 100082 100083 100084 100090 100091 100092 100093 100103 100104 100105 100106 100107
x 457248.9 457162.2 456996.1 453751.9 453780.3 453752.7 451032.5 450975.5 451679.3 455508.5 455369 455176.6 458700.6 459015.6 458960.8 455508.5 455798.3 458700.6 459159.3 453751.7 455176.5 457248.9 458960.8 450975.5 453248.6 453752.7 456631.3 456996.2
y 4236825 4233376 4230205 4236812 4233170 4229944 4233596 4229819 4237023 4244104 4240783 4237279 4243995 4240649 4237100 4244104 4242871 4243995 4242646 4236812 4237280 4236825 4237100 4229819 4230911 4229944 4231286 4230205
z z-dem 297.866 298 69.651 74 213.421 217 200.879 198 47.608 50 36.865 45 35.523 38 3.13 14 118.044 110 81.211 91 220.329 220 289.028 297 173.146 173 366.994 365 401.704 395 81.333 91 109.029 108 173.114 173 322.849 320 200.981 198 289.402 297 298.244 298 401.592 395 3.106 14 55.988 59 36.846 45 54.338 43 213.048 217 mean error RMSE
103
dz -0.134 -4.349 -3.579 2.879 -2.392 -8.135 -2.477 -10.87 8.044 -9.789 0.329 -7.972 0.146 1.994 6.704 -9.667 1.029 0.114 2.849 2.981 -7.598 0.244 6.592 -10.894 -3.012 -8.154 11.338 -3.952 4.936 6.022
APPENDIX E
104
105
106
107
108
109
110
111
112
APPENDIX F
Table F.1 Point feature measurements on orthoimage generated using DEM from SPOT with automatic correlation in PCI
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
ref-x 452079.3 455914.3 454717.5 453808.3 457086.5 451779.3 455184.3 457342.8 458345.5 456473.3 453745.5 454596.5 457708.5 457890.5 455328.5 453423.5 455050.3 458786.3 457681.5 456209.5 454500.5 453830.3 455301.5 456808.5 458768.5 458843.5 457381.5 455640.5 453416.5 453744.8 454685.5 455421.5 456639.5 457496.5 452688.5 452722.5 454951.5 457327.5 454599.5 452402.5 456125.5 456553.5 454855.5 455221.5 451351.5
ref-y 4229863.8 4229766.8 4232836.5 4232484.3 4233133.5 4236350.3 4236272.3 4235248.3 4238211.5 4238690.3 4238937.5 4239453.5 4240559.5 4243234.5 4241463.5 4240171.5 4244191.8 4241375.8 4243024.5 4243211.5 4244366.5 4242022.3 4242043.5 4241702.5 4241552.5 4239574.5 4239205.5 4239635.5 4240172.5 4238936.3 4236286.5 4236364.5 4235682.5 4235476.5 4235093.5 4237322.5 4237597.5 4236720.5 4234876.5 4233086.5 4233932.5 4231652.5 4230374.5 4231591.5 4229804.5
ort-x 452060.4 455911.6 454707.9 453799.1 457084.1 451778.4 455174.1 457340.4 458350.4 456476.6 453746.6 454592.6 457706.6 457882.3 455337.3 453427.9 455030.4 458782.1 457690.7 456196.3 454505.4 453816.3 455282.9 456801.6 458770.4 458844.1 457374.1 455649.1 453422.9 453744.1 454694.1 455421.6 456632.9 457500.4 452682.9 452719.1 454942.9 457325.4 454595.4 452402.9 456131.6 456555.4 454861.6 455237.9 451347.9
ort-y 4229840.1 4229778.9 4232837.6 4232482.6 4233138.9 4236351.3 4236275.1 4235243.9 4238212.6 4238677.6 4238928.9 4239456.7 4240565.1 4243238.3 4241465.8 4240168.9 4244197.6 4241375.1 4243026.1 4243212.3 4244372.6 4242002.9 4242028.9 4241703.9 4241546.4 4239575.1 4239206.4 4239636.4 4240163.9 4238932.6 4236287.6 4236370.1 4235682.6 4235473.9 4235092.6 4237320.2 4237593.9 4236721.4 4234873.9 4233080.9 4233928.9 4231653.9 4230368.9 4231583.9 4229811.4 mean error RMSE
113
dx 18.875 2.625 9.625 9.125 2.375 0.875 10.125 2.375 -4.875 -3.375 -1.125 3.938 1.875 8.250 -8.750 -4.375 19.875 4.175 -9.188 13.188 -4.875 13.938 18.625 6.875 -1.875 -0.625 7.375 -8.625 -6.375 0.625 -8.625 -0.125 6.625 -3.875 5.625 3.375 8.625 2.125 4.125 -0.375 -6.125 -1.875 -6.125 -16.375 3.625 2.029 7.944
dy 23.625 -12.125 -1.125 1.625 -5.375 -1.000 -2.875 4.375 -1.125 12.625 8.625 -3.188 -5.625 -3.750 -2.250 2.625 -5.875 0.625 -1.562 -0.812 -6.125 19.312 14.625 -1.375 6.125 -0.625 -0.875 -0.875 8.625 3.625 -1.125 -5.625 -0.125 2.625 0.875 2.350 3.625 -0.875 2.625 5.625 3.625 -1.375 5.625 7.625 -6.875 1.522 6.754
ds 30.239 12.406 9.691 9.269 5.876 1.329 10.525 4.978 5.003 13.068 8.698 5.067 5.929 9.062 9.035 5.102 20.725 4.222 9.320 13.213 7.828 23.816 23.681 7.011 6.406 0.884 7.427 8.669 10.725 3.678 8.698 5.626 6.626 4.680 5.693 4.113 9.356 2.298 4.889 5.637 7.117 2.325 8.316 18.063 7.772 8.758 6.071
Table F.2 Point feature measurements on orthoimage generated using DEM from 1:25.000 scale contour lines
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
ref-x 452079.3 455914.3 454717.5 453808.3 457086.5 451779.3 455184.3 457342.8 458345.5 456473.3 453745.5 454596.5 457708.5 457890.5 455328.5 453423.5 455050.3 458786.3 457681.5 456209.5 454500.5 453830.3 455301.5 456808.5 458768.5 458843.5 457381.5 455640.5 453416.5 453744.8 454685.5 455421.5 456639.5 457496.5 452688.5 452722.5 454951.5 457327.5 454599.5 452402.5 456125.5 456553.5 454855.5 455221.5 451351.5
ref-y 4229863.8 4229766.8 4232836.5 4232484.3 4233133.5 4236350.3 4236272.3 4235248.3 4238211.5 4238690.3 4238937.5 4239453.5 4240559.5 4243234.5 4241463.5 4240171.5 4244191.8 4241375.8 4243024.5 4243211.5 4244366.5 4242022.3 4242043.5 4241702.5 4241552.5 4239574.5 4239205.5 4239635.5 4240172.5 4238936.3 4236286.5 4236364.5 4235682.5 4235476.5 4235093.5 4237322.5 4237597.5 4236720.5 4234876.5 4233086.5 4233932.5 4231652.5 4230374.5 4231591.5 4229804.5
ort-x 452060.4 455920.4 454706.6 453805.4 457086.6 451774.1 455181.6 457342.9 458350.4 456476.6 453745.4 454596.5 457706.6 457889.8 455337.3 453431.6 455040.4 458792.9 457686.9 456211.3 454501.6 453820.7 455287.9 456802.9 458774.1 458850.4 457380.4 455645.4 453419.1 453745.6 454699.1 455420.4 456639.8 457505.4 452680.4 452720.4 454951.6 457327.9 454600.4 452402.9 456132.9 456556.6 454851.6 455235.4 451350.4
ort-y 4229837.4 4229779.9 4232838.6 4232481.1 4233137.4 4236343.6 4236276.1 4235242.4 4238214.9 4238678.6 4238937.4 4239457.8 4240564.9 4243241.8 4241466.8 4240167.4 4244197.4 4241373.6 4243027.7 4243215.2 4244373.6 4242002.1 4242028.6 4241704.9 4241546.1 4239569.9 4239206.1 4239638.6 4240167.4 4238932.6 4236284.9 4236367.4 4235684.3 4235471.1 4235092.4 4237319.9 4237591.1 4236718.6 4234868.6 4233079.9 4233936.1 4231652.4 4230376.1 4231579.9 4229813.6 mean error RMSE
114
dx 18.875 -6.125 10.875 2.875 -0.125 5.125 2.625 -0.125 -4.875 -3.375 0.125 0.031 1.875 0.750 -8.750 -8.125 9.875 -6.625 -5.438 -1.812 -1.125 9.562 13.625 5.625 -5.625 -6.875 1.125 -4.875 -2.625 -0.875 -13.625 1.125 -0.250 -8.875 8.125 2.125 -0.125 -0.375 -0.875 -0.375 -7.375 -3.125 3.875 -13.875 1.125 -0.376 6.632
dy 26.375 -13.125 -2.125 3.125 -3.875 6.625 -3.875 5.875 -3.375 11.625 0.125 -4.344 -5.375 -7.250 -3.250 4.125 -5.625 2.125 -3.188 -3.688 -7.125 20.188 14.875 -2.375 6.375 4.625 -0.625 -3.125 5.125 3.625 1.625 -2.875 -1.750 5.375 1.125 2.625 6.375 1.875 7.875 6.625 -3.625 0.125 -1.625 11.625 -9.125 1.527 7.465
ds 32.433 14.484 11.081 4.246 3.877 8.376 4.680 5.876 5.929 12.105 0.177 4.344 5.693 7.289 9.334 9.112 11.365 6.957 6.304 4.109 7.213 22.338 20.172 6.106 8.502 8.286 1.287 5.791 5.758 3.729 13.722 3.087 1.768 10.376 8.203 3.377 6.376 1.912 7.923 6.636 8.218 3.127 4.202 18.101 9.194 8.070 5.969
Table F.3 Point feature measurements on orthoimage generated using DEM from aerial photographs with automatic correlation in DPW
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
ref-x 452079.3 455914.3 454717.5 453808.3 457086.5 451779.3 455184.3 457342.8 458345.5 456473.3 453745.5 454596.5 457708.5 457890.5 455328.5 453423.5 455050.3 458786.3 457681.5 456209.5 454500.5 453830.3 455301.5 456808.5 458768.5 458843.5 457381.5 455640.5 453416.5 453744.8 454685.5 455421.5 456639.5 457496.5 452688.5 452722.5 454951.5 457327.5 454599.5 452402.5 456125.5 456553.5 454855.5 455221.5 451351.5
ref-y 4229863.8 4229766.8 4232836.5 4232484.3 4233133.5 4236350.3 4236272.3 4235248.3 4238211.5 4238690.3 4238937.5 4239453.5 4240559.5 4243234.5 4241463.5 4240171.5 4244191.8 4241375.8 4243024.5 4243211.5 4244366.5 4242022.3 4242043.5 4241702.5 4241552.5 4239574.5 4239205.5 4239635.5 4240172.5 4238936.3 4236286.5 4236364.5 4235682.5 4235476.5 4235093.5 4237322.5 4237597.5 4236720.5 4234876.5 4233086.5 4233932.5 4231652.5 4230374.5 4231591.5 4229804.5
ort-x 452061.9 455931.9 454706.9 453805.6 457091.9 451779.4 455186.9 457343.1 458349.4 456476.9 453745.6 454595.5 457710.6 457888.8 455333.8 453433.1 455038.1 458789.4 457685.3 456210.3 454501.9 453821.6 455293.1 456801.9 458773.1 458848.1 457383.1 455648.1 453419.4 453745.4 454701.9 455423.1 456635.6 457505.6 452683.1 452720.6 454954.3 457330.6 454596.9 452403.1 456133.1 456554.4 454855.6 455235.6 451361.9
ort-y 4229836.9 4229773.1 4232836.9 4232484.4 4233138.1 4236340.6 4236276.9 4235239.4 4238216.9 4238675.6 4238935.6 4239454.5 4240568.1 4243236.3 4241461.3 4240168.1 4244198.1 4241373.1 4243034.1 4243210.3 4244371.9 4242002.8 4242028.1 4241703.1 4241545.6 4239569.4 4239199.6 4239638.1 4240165.6 4238936.1 4236288.1 4236363.1 4235680.6 4235469.4 4235093.1 4237318.1 4237593.1 4236718.1 4234874.4 4233080.6 4233926.9 4231651.9 4230375.6 4231584.4 4229810.6 mean error RMSE
115
dx 17.375 -17.625 10.625 2.625 -5.375 -0.125 -2.625 -0.375 -3.875 -3.625 -0.125 1.031 -2.125 1.750 -5.250 -9.625 12.125 -3.125 -3.821 -0.812 -1.375 8.688 8.375 6.625 -4.625 -4.625 -1.625 -7.635 -2.875 -0.625 -16.375 -1.625 3.875 -9.125 5.375 1.875 -2.825 -3.125 2.625 -0.625 -7.625 -0.875 -0.125 -14.125 -10.375 -1.461 6.904
dy 26.875 -6.375 -0.375 -0.125 -4.625 9.625 -4.625 8.875 -5.375 14.625 1.875 -1.031 -8.625 -1.750 2.250 3.375 -6.375 2.625 -9.562 1.188 -5.375 19.438 15.375 -0.625 6.875 5.125 5.925 -2.625 6.875 0.125 -1.625 1.375 1.875 7.125 0.375 4.375 4.375 2.375 2.125 5.875 5.625 0.625 -1.125 7.125 -6.125 2.399 7.182
ds 32.002 18.742 10.632 2.628 7.091 9.626 5.318 8.883 6.626 15.068 1.879 1.458 8.883 2.475 5.712 10.200 13.699 4.081 10.297 1.439 5.548 21.291 17.508 6.654 8.286 6.903 6.144 8.074 7.452 0.637 16.455 2.129 4.305 11.577 5.388 4.760 5.208 3.925 3.377 5.908 9.475 1.075 1.132 15.820 12.048 8.174 6.245
Table F.4 Point feature measurements on orthoimage generated using DEM from SPOT with automatic correlation in DPW
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
ref-x 452079.3 455914.3 454717.5 453808.3 457086.5 451779.3 455184.3 457342.8 458345.5 456473.3 453745.5 454596.5 457708.5 457890.5 455328.5 453423.5 455050.3 458786.3 457681.5 456209.5 454500.5 453830.3 455301.5 456808.5 458768.5 458843.5 457381.5 455640.5 453416.5 453744.8 454685.5 455421.5 456639.5 457496.5 452688.5 452722.5 454951.5 457327.5 454599.5 452402.5 456125.5 456553.5 454855.5 455221.5 451351.5
Ref-y 4229863.8 4229766.8 4232836.5 4232484.3 4233133.5 4236350.3 4236272.3 4235248.3 4238211.5 4238690.3 4238937.5 4239453.5 4240559.5 4243234.5 4241463.5 4240171.5 4244191.8 4241375.8 4243024.5 4243211.5 4244366.5 4242022.3 4242043.5 4241702.5 4241552.5 4239574.5 4239205.5 4239635.5 4240172.5 4238936.3 4236286.5 4236364.5 4235682.5 4235476.5 4235093.5 4237322.5 4237597.5 4236720.5 4234876.5 4233086.5 4233932.5 4231652.5 4230374.5 4231591.5 4229804.5
ort-x 452063.1 455918.1 454714.4 453798.1 457090.6 451770.6 455181.9 457340.6 458355.6 456478.1 453746.9 454589.8 457711.9 457891.3 455331.3 453433.1 455036.9 458791.9 457690.3 456205.3 454498.1 453815.9 455286.9 456800.6 458773.1 458848.1 457378.1 455646.9 453421.9 453745.9 454695.6 455424.4 456639.4 457500.6 452666.9 452719.4 454954.4 457325.6 454596.9 452410.6 456133.1 456558.1 454853.1 455236.9 451345.6
ort-y 4229834.9 4229778.6 4232837.4 4232483.6 4233138.6 4236349.9 4236274.9 4235241.1 4238212.4 4238677.4 4238937.4 4239457.5 4240566.1 4243236.8 4241461.8 4240171.1 4244198.6 4241372.4 4243027.1 4243215.8 4244371.1 4242002.7 4242031.1 4241702.4 4241543.6 4239569.9 4239196.1 4239638.6 4240163.6 4238935.9 4236291.1 4236363.6 4235682.4 4235469.9 4235093.6 4237319.9 4237592.4 4236722.4 4234873.6 4233084.9 4233927.4 4231649.9 4230374.9 4231579.9 4229814.9 mean error RMSE
116
dx 16.125 -3.875 3.125 10.125 -4.125 8.625 2.375 2.125 -10.125 -4.875 -1.375 6.656 -3.375 -0.750 -2.750 -9.625 13.375 -5.625 -8.812 4.188 2.375 14.312 14.625 7.875 -4.625 -4.625 3.375 -6.375 -5.375 -1.188 -10.125 -2.875 0.125 -4.125 21.625 3.125 -2.875 1.875 2.625 -8.125 -7.625 -4.625 2.375 -15.375 5.875 0.303 7.880
dy 28.875 -11.875 -0.875 0.625 -5.125 0.375 -2.625 7.125 -0.875 12.875 0.125 -4.031 -6.625 -2.250 1.750 0.375 -6.875 3.375 -2.562 -4.312 -4.625 19.562 12.375 0.125 8.875 4.625 9.375 -3.125 8.875 0.312 -4.625 0.875 0.125 6.625 -0.125 2.625 5.125 -1.875 2.875 1.625 5.125 2.625 -0.375 11.625 -10.375 1.905 7.445
ds 33.072 12.491 3.245 10.144 6.579 8.633 3.540 7.435 10.163 13.767 1.381 7.781 7.435 2.372 3.260 9.632 15.038 6.560 9.177 6.011 5.199 24.239 19.158 7.876 10.008 6.541 9.964 7.100 10.376 1.228 11.131 3.005 0.177 7.804 21.625 4.081 5.876 2.652 3.893 8.286 9.187 5.318 2.404 19.275 11.923 8.801 6.490
APPENDIX G Table G.1 Line feature measurements on orthoimages ds x
y 451531 451700 451941 452120 452307 452441 452611 452741 453031 453311 453591 453771 453811 453711 453711 454011 454301 454521 454961 455361 455621 456231 456451 456831 456230 456441 456461 456521 456581 452121 451847
(m)
1 :25.000 Aerial Photo. SPOT (PCI) SPOT (DPW) 4229779 8.50 8.75 5.00 5.00 4229774 8.75 7.00 6.00 6.00 4229797 8.75 3.75 8.00 5.75 4229882 7.50 10.00 14.00 14.00 4230047 8.75 5.25 1.25 5.00 4230412 5.00 9.25 1.25 5.00 4230463 5.00 4.25 5.00 2.50 4230653 0.00 15.00 0.00 11.00 4230763 7.00 12.00 3.75 12.00 4230813 0.00 1.25 0.00 2.00 4230953 6.00 15.00 3.75 6.00 4231062 2.50 4.00 5.00 3.75 4231373 6.25 4.50 10.50 2.50 4231663 0.00 5.00 2.50 5.00 4232023 2.00 2.00 4.00 10.00 4232262 1.00 5.50 8.75 16.00 4232493 3.00 7.50 5.00 13.50 4232683 2.50 1.00 3.00 8.00 4232853 1.00 3.75 0.50 1.25 4232919 2.00 0.50 2.00 2.50 4232914 1.25 3.75 3.75 1.25 4233243 4.00 1.25 7.00 3.75 4233400 5.50 0.00 2.50 1.25 4233443 6.50 0.50 3.00 6.50 4229793 7.50 3.00 8.00 4.00 4230713 5.00 6.00 1.50 2.50 4231113 1.25 3.75 11.50 3.75 4231453 6.25 0.00 10.00 15.00 4231743 3.00 4.00 5.00 9.00 4235643 2.50 6.50 5.00 10.50 4236075 0.00 6.25 1.25 3.50
117
451751 451591 451030 451481 451871 451181 452561 453571 453841 454181 454531 454841 455221 455571 455961 456321 456731 457090 457431 457881 458161 458371 452871 453141 453481 453711 454001 454421 453531 453652 454281 454641 454841 455151 455451
4236473 4236783 4237216 4237423 4237643 4237893 4238283 4238893 4239053 4239313 4239463 4239633 4239758 4239763 4239706 4239642 4239543 4239383 4239213 4238903 4238473 4238193 4240443 4240283 4240147 4239963 4239786 4239785 4240253 4240593 4240813 4241003 4241133 4241203 4241443 mean error RMSE
14.50 4.50 2.00 7.00 10.50 14.00 8.00 3.75 10.00 3.25 11.00 7.50 4.00 8.75 3.00 10.00 3.75 2.50 3.25 3.50 15.00 7.50 4.50 5.00 2.50 3.75 3.00 3.50 8.75 2.00 6.25 7.50 2.50 4.00 1.50 5.155 3.520
2.50 2.00 1.00 0.50 2.50 8.00 0.50 12.50 2.00 3.00 2.00 15.00 5.00 6.25 4.50 12.00 10.00 3.75 13.25 6.25 6.25 14.00 4.50 3.50 1.00 2.50 0.00 0.00 1.00 2.75 6.25 1.25 1.25 6.25 2.50 4.936 4.125
118
9.00 2.50 2.25 7.50 6.50 5.00 4.00 15.00 6.50 2.50 12.50 14.00 5.00 0.50 6.00 1.25 12.00 0.00 9.00 6.25 16.00 17.50 2.75 7.50 1.00 10.00 1.50 1.25 7.00 0.50 8.50 6.25 1.00 2.50 4.00 5.489 4.277
11.00 1.25 1.25 3.75 5.75 6.25 3.75 0.50 6.00 2.75 2.50 10.00 5.50 6.00 18.25 3.75 0.00 2.00 2.50 10.00 17.00 5.50 8.00 1.25 5.00 14.00 12.50 2.50 7.00 0.50 5.00 4.00 1.50 7.00 2.00 5.958 4.472
APPENDIX H
Table H.1 Area measurements of area features on orthoimages (m2)
area
d-area (m2)
Reference 1 :25.000 Aerial Pho. SPOT (PCI) SPOT (DPW) 1 :25.000 Aerial Pho. SPOT (PCI) SPOT (DPW)
1 2 3 4 5 6 7 8
28434 9162 6836 11900 6674 28640 44438 3705
28415 8195 7087 10192 6962 27370 42586 3604
27139 8429 6594 9302 6794 27234 40159 3537
26397 8483 5747 8915 5488 30695 41526 3018
26760 19 1295 7857 967 733 5890 -251 242 9687 1708 2598 6273 -288 -120 26190 1270 1406 40687 1852 4279 3046 101 168 mean error 807.00 1355.12 RMSE 824.34 1383.12
2037 679 1089 2985 1186 -2055 2912 687 1703.75 1498.29
1674 1305 946 2213 401 2450 3751 659 1674.88 1032.20
Table H.2 Perimeter measurements of area features on orthoimages perimeter (m)
d-perimeter (m)
Reference 1 :25.000 Aerial Pho. SPOT (PCI) SPOT (DPW) 1 :25.000 Aerial Pho. SPOT (PCI) SPOT (DPW)
1 2 3 4 5 6 7 8
685 387 346 433 333 778 1069 292
683 372 339 405 346 787 1044 282
670 374 328 389 336 788 1028 281
671 367 305 384 305 794 1045 260
661 358 313 398 326 778 1023 263 mean error RMSE
119
2 15 7 28 -13 -9 25 10 13.62 14.68
15 13 18 44 -3 -10 41 11 19.38 18.84
14 20 41 49 28 -16 24 32 28.00 19.68
24 29 33 35 7 0 46 29 25.38 15.05
H e i g h t E r r o r s (m)
Figure E.1 Error profile of DEM produced from aerial photographs with automatic correlation in DPW in North-South direction
H e i g h t E r r o r s (m)
Figure E.2 Error profile of DEM produced from SPOT in DPW in North-South direction
H e i g h t E r r o r s (m)
Figure E.3 Error profile of DEM produced from SPOT in PCI in North-South direction
H e i g h t E r r o r s (m)
Figure E.4 Error profile of DEM produced from aerial photographs with automatic correlation in DPW in East-West direction
H e i g h t E r r o r s (m)
Figure E.5 Error profile of DEM produced from SPOT in DPW in East-West direction
H e i g h t E r r o r s (m)
Figure E.6 Error profile of DEM produced from SPOT in PCI in East-West direction
H e i g h t E r r o r s (m)
Figure E.7 Error profile of DEM produced from aerial photographs with automatic correlation in DPW along a road
H e i g h t E r r o r s (m)
Figure E.8 Error profile of DEM produced from SPOT in DPW along a road
H e i g h t E r r o r s (m)
Figure E.9 Error profile of DEM produced from SPOT in PCI along a road
