POLITECNICO DI TORINO

DIATI

Photogrammetry – part 3
Plotting, image data processing, digital
products
(Section 2)
MARCO PIRAS
GEOMATICS: 01RVUMX
 Feature Extraction
Forstner’s operatot
Forstner operator is one of the most used photogrammetric teqchniques
devoted to extract the interesting points.
This operator is based on LSM (Least Square Matching) , applied on a two-
dimensional translation
g 1 x , y   c  g 2 x  t x , y  t y   b
Considering a size matrix = n.n around a pixel in a digital image, it is possible to
demonstrate that variance-covariane Cxx of the LSM system is:

 ... ... ... ... 

  c2 ... ... ...   
   ... ... ... ... 
 ...  2
... ...  1  ... ...
C xx  ˆ 02 N 1   C xx  ˆ 02 N 1   g 2 '2y   g 2 'x g 2 ' y 
b
2 
 ... ...  tx  txty  det(N)  
2

 ... ...  g 2 '2x 

nxn nxn
 ... ...  2
 tytx  ty2 

 g ' g2 'y
2 x  
 ... ... ... ...  nxn nxn

 
 ... ... ... ...  g 2' x  g 2 x1,y
 g 2 x, y
N   ...  g 2 '2x  g 2 'x g 2 ' y 

...
 g 2' y  g 2 x, y1
 g 2 x, y
 ... g 2 '2y 
nxn nxn



... g
nxn
2 x ' g2 'y nxn 
 Feature Extraction
Forstner operator
The Forstner operator selects points of interest by analyzing
the size and shape of the standard error ellipses provided by
the Cxx variance and covariance matrix of the LSM equation
system

The size of the ellipses is defined by w. The shape of the

ellipses is defined by q, so called circularity parameter:

4 det N x,y 
q
tr N x,y 
2

If q=0 N is singular. The point is a radiometric edge.

If q=1 the ellipse is a circle. This is a point of interest, which
has a high radiometric contrast with respect the closest
points
 Feature Extraction
Forstner’s operator
examples

Forstner operator: w and q

are setting by the user.

1300 points, camera DMC, GSD= 34 cm, 600 dpi)

 Image Matching
Feature Based Matching
In FBM techniques, automatic matching between homologous points occurs by comparing a
set of attributes that characterize the radiometric edge of each single point (feature).
For each feature of each image is associated an n-dimensional vector, called "descriptor"
(feature descriptor) that represents and describes synthetically the radiometric
characteristics in its surroundings.

The matching is done by evaluating the degree of similarity between the descriptors. The
candidate points to the matching can be defined as homologous points if the Euclidean
distance between the two descriptors is minimal.

SIFT descriptor (Scale Invariant Feature Transform, Lowe 2004)

Euclidean distance
N
 
 D 
32

 min j   
2
DK DNN Ki D ji

1
i 1 

n=32
 Image Matching
Feature Based Matching

The main advantage of the FBM algorithms is the capability to automatically coupling the
homologous points belonging to stereoscopic images with strong geometrical variations
(scale, rotations, perspective distortions) and radiometric ones (lighting conditions, noise
etc.) even ensuring a good accuracy (0.5- 1 pixel).

The disadvantage of these

techniques is due to the
complexity of the
algorithms and the high
computational time
 Automatic photogrammetric procedures
Combining the different algorithms, it is possible to obtain automatic or semi-automatic
photogrammetric process.
Manual operations
Autom. operation
PHOTOGRAMMETRIC ACQUISITION
Autom. algorithm

Pre-processing of images HP, LP filtering

ORIENTATIONS
Inner orientation CC - LSM

Tie points extraction Forstner - CC - LSM

GCPs
DETERMINATION Aerial triangulation LMS

CARTOGRAPHIC PRODUCTS

DSM extraction Forstner - CC - LSM

CC-LSM Plotting
orthoprojection Resampling
 Automatic photogrammetric procedures

The automation in digital photogrammetry needs to learn new concepts:

- The determination of approximate parameters for a non-linear systems becomes
a more important problem, because direct solutions are often required (linear)
- In order to avoid the manual activities, robust estimators have a fundamental role
due to the presence in the data automatically acquired (autocorrelation) of a large
number of accidental, systematic and coarse errors

The automatic photogrammetric procedures must have two important properties:

- Robustness as the final solution is not influenced (or the influence is very very
low) by model errors
- Diagnostics as the ability to identify the model errors
 Digital mapping using images
summary

Rectification
Ortho-rectified image
Precise orthophoto
Solid Orthophotos
 The photomapping
Definition
Photogrammetry is a technique that produces photomapping, using the
radiometric information of the acquired images. These products integrates the
information provided by traditional map or numerical ones (vector files)

A photomapping is defined as a map,

which is metrically correct, where the
horizontal component of the
coordinates (X,Y) is represented as an
image, derived from one or more aerial
photos of the area

In order to represent the vertical component, a photomapping can be integrated with

a vertical known points, contour lines and toponym.
 The photomapping
Definition
In an aerial case, the photomappping is quite similar to a frame (photogram)

Frame??
? Photomapping??
 The photomapping
Photogram deformation
The difference between photogram and the photomapping is mainly a geometric
difference:
- PHOTOGRAM is a perspective representation (central perspective) of the object
(the terrain in aerial photogrammetry)
- PHOTOMAPPING is a cartographic representation, on a predefined scale factor, of
the same object  Y

3' 4' 3 4

1' 2' 1 2

 X
The central perspective of a regular grid produces an irregular and deformed grid.
The example shown in the figure represents schematically the geometric
deformation effect for each frame.
 The photomapping
Photogram deformation
 Y
The values of the
deformation can be
3' 4' 3 4 estimated using , and
X, Y (terrain coordinates).
1' 2' 1 2

 X

Length deformation
  These deformations are different on each
 X  1 2  Y  13 point of the image and depends on the
X Y
direction. It is similar to the cartographic
dove ik  i   k 2  i  k 2 deformation
The central perspective could be
Right angle deformation
considered as a generic stereographic
100  100 gon  3'1'2' azimuthal projection
Area deformation

S   X   Y  cos 100
 Rectification
Definition
If the surface of the object is a plane, the centrar perspective is like the object in the
following figure and the relation between terrain points and image points are described
by a general equation of the homography.
 Y a1  a2  a3
X
c1  c2  1

b1  b2  b3
Y
c1  c2  1

 X
If there are the hypotheses of planarity of the object, the deformations of the
frame can be modeled with the equations of general homography. The operation
eliminates these deformations and it is called RECTIFICATION.

In this case, the frame is transformed into a central projection of the object, with
a plane parallel to the object itself and principal axis perpendicular to the object
plane. This new central perspective is the identical to an orthogonal projection
of the object, excluding a scale factor .
Rectification
Definition
 Rectification
Procedure
Giving the image of the object and a series of
known points (X,Y) on the object:
- homography parameters are estimated (at ORIGINAL IMAGE
least 4 known points)
- A rectified image is considered (at the
begin it is empty matrix) , where each pixel
is defined with the object coordinates X, Y
COLOR
- For each pixel in the image, homography
equation is applied, with purpose to project X,Y -> ,
Z
the object point into image space. HOMOGRAPHY

- radiometric value is resampling

(RESAMPLING) and it is assigned to this Y
RECTIFIED IMAGE
pixel
- this procedure is repeated for all pixels, in X

order to obtain the RECTIFIED IMAGE X  X0  d * ( c  1 )

Y  Y0  d * ( r  1 )
 Rectification
The systematic error in the height
What are the height variation which can be considered as negligible in the
rectification? Which parameters do they depend on?
O
The image points that do not
c exactly belong to the object plane
are translated in the direction of the
 PP  principal point.
  The value of this radial
Z
0 displacement, in the case of
normal image, is expressed by
the following relation:

c  
  R  Z  Z
R Z Z0 Z0 c  mb
Z R

This relation is also valid in the case of photogram collected in «quasi» normal
condition, with a good approximation.
 Rectification
The systematic error in the height
Considering the previous equation, we can made these remarks:

• In the rectification, it is important to use only the central part of the

frame;
• it is better to use lens with hight focal distance (c)

When the portion of the image and the principal distance are defined, it is
possible to estimate what is the maximum value of displacement from the object
plane, in order to have the error in height negligible with respect to the scale of
the map

  c  mb Rmax  c Rmax  2  0.2  0.6 mm  scala

Scale of
Z max   fotopiano
the photomapping
 
 Rectification
Scala
Scale Scala
Scale c = 45 c = 100 c = 200 mm
map
fotopiano frame
fotogramma
1:50 1:200 3 6 12 cm
1:400 5 11 24 cm
1:100 1:400 5 11 24 cm
1:800 10 23 48 cm
1:250 1:1.000 13 28 59 cm
1:2.000 25 57 119 cm

Scala
Scale Scala
Scale c = 90 c = 150 c = 210 c = 300 mm
map frame
fotopiano fotogramma
1:1.000 1:3.000 1 1 2 3m
1:6.000 2 3 4 5m
1:2.500 1:7.500 2 3 4 6m
1:15.000 4 6 9 13 m
1:5.000 1:15.000 4 6 9 13 m
1:30.000 8 13 18 26 m
1:10.000 1:30.000 8 13 18 26 m
1:60.000 15 26 36 51 m
Rectification
Examples
 Orthoprojection
Definition
If the object is not a plane, the homography equations are not more sufficient to
eliminate the deformations of the frame.
It is necessary to work with the orthprojection. This technique allows the
geometrically transformation of an image into an image that can be considered as a
cartographic or orthogonal projection of the object.
Orthoprojection is a technique that requires prior knowledge of the shape of the
object which geometrically means knowing one of the three coordinates of the
points.

Using digital image, the

orthoprojection approach is quite
changed.
Nowadays, the procedure is made
by specific software, which allow to
reduce the cost and to extend the
number of application fields.
 Orthoprojection
Definition

The orthophoto is a cartographic

   ( X , Y , Z / par ) representation defined by a geo-
   ( X , Y , Z / par ) referenced digital image: it corresponds
to an orthogonal projection of the
terrain in the cartographic plane.
For each pixel:
• the horizontal componets are
defined(E,N)
• using a terrain model, an interpolation
of (Z) is made
• the 3D point is projected on the image
(,) with a dedicated model
•DN (digital number) is estimated and it
is applied in the image position with
resampling
•DN of the pixel is stored in the OP
 Orthoprojection
Classification
There are three different types of orthophotos:
• Rapid OP
• Normal OP
• Precise OP

Rapid orthophoto is a photographic representation, which is produced with

rapid and cheaper techniques. The graphical precision is 0.3 mm of the
nominal scale of the orthophoto for the objects on the terrain, and 0.9 mm of
the nominal scale for all objects with a elevation ( building, bridge, viaduct).
It is generally used for environmental application or thematic map (emergency
management or early mapping).

The normal orthophoto is a photographic representation, but with 0.2 nominal

scale of the orthophoto for the objects on the terrain, and 0.6 mm of the
nominal scale for all objects with a elevation.
 Orthoprojection
Normal orthophoto
The orthoprojection is realized using a digital terrain model (DTM).
C
IQ

Q0
P0 Q

DTM P

orthophoto
P0 Q0

This product has some residual

perspective deformations, due to a not
correct geometry model.
 Orthoprojection
Precise orthophoto
In this case, all perspective deformations are
removed, using a rigorous model !

Precise OP
 L’ortofoto di precisione
Definizione

It this case, the precision is 0.2mm of the nominal scale of the

orthophoto all objects, even in the elevation.
 Precise orthophoto
Geometry (shape) model
In order to produce a precise ORTHOPHOTO, it is important to use a correct
geometry (shape) model. This model could be created by:

1. A digital surface model (DSM) , starting by a

digital terrain model (DTM). MANUAL Y
PROCEDURE and SHAPE FORMAT

2. Dense digital surface model (DDSM) in

RASTER FORMAT. It is a model with an
high density of points
AUTOMATIC PROCEDURE.
It is generated usign numerical map, LIDAR
or photogrammetry
 Precise orthophoto
Images duplication
The correct shape model (DDSM) is not sufficient to eliminate the perspective
deformations. In fact, using a correct description of the shape of the object and a single
image, some visible points are duplicated on the orthophotos
C

P0
IQ

Q0 Q

DDSM
P

Orthophoto

Q0 P0
 Precise Orthophoto
How to procude the Precise OP

DDSM is fundamental to
generate a precise OP.

C1 C2

IP IQ

P DDSM
Several images (multi-image
approach), in order to have
orthophoto complete photographic
P0 Q0 description of the object.
 Precise orthophoto
How to produce a Precise OP

QP QR QR QS FLAG

I2 I1

R S

P Q
 Precise orthophoto
A terrestrial case

DSM (Laser
Scanner)

images

Camera (Rollei 6008)

 Precise orthophoto
A terrestrial case

First result
Precise orthophoto
A terrestrial case
 Precise orthophoto
A terrestrial case

Final result
 Precise orthophoto
An Aerial case

Orthophoto of Turin
municipality

INPUT
Numerical map 1:2000
Photogrammetric flight frame scale
1:5000.
 L’ortofoto di precisione
An aerial case
Generation of a DDSM from a numerical map

resolution: 20 cm
 Precise orthophoto
An aerial case

GSD: 20 cm
 Precise orthophoto
An aerial case

S.S. Consolata
Gran Madre

Duomo, Cupola della

Sindone, Palazzo Reale
 Orthophoto in large scale map
CISIS: italian guideline
 Precise solid orthophoto
Definition

PSO is the fusion of the two-

dimensional and chromatic
content of the precision
Z orthophoto (OP) with the three-
dimensional information of a
DDSM (Dense Digital Surface
Model) of object.
X
The position of the selected
orthophoto pixel gives the
Y
horizontal coordinates while the
altimetric model contains the
corresponding height.
x  x0  d * (c  1)
y  y0  d * (r  1)
z  q (c, r )
 Precise solid orthophoto
Structure

height DDSM
1. Hsi, 2 byte/pixel (short integer) 2-4 byte/pixel
2. Hf, 4 byte/pixel (float) bil,bip,bsq

(Red)
1 byte/pixel (unsigned character) orthophoto
(Green) 3 byte/pixel
1 byte/pixel (unsigned character) TIF+TFW
(Blue)
1 byte/pixel (unsigned character)

For each additional band

1 byte/pixel (unsigned character) Other
bands
1 byte/pixel
bil,bip,bsq
 Final remarks
The use of modern geomatics techniques (digital photogrammetry, LiDAR, high
resolution images) gives the opportunity to define, in a rapid and economically
sustainable way, complex and complete three-dimensional representations of
the objects in the environmental, architectural, engineering and geoengineering
fields.

The specific use of these techniques must be done (using the brain!!) by
experts of data acquisition, process and able to generate the final products.

The solid image and the precise solid orthophoto allow the visualization of 3D
data , with the following characteristics:
- Simple (standard PC, no specific equipment)
- Efficient (allows to obtain all information)
- Rapid (it does not require specific preparation with specific SW)

These products are optimal in multidisciplinary approaches because they allow

to share the same data to different user with different competences.