).
of
St
ta
11
al
nt
OT
ccurate
ful for
These
osition
pment
rst, the
ice the
'ussed.
tation
ussion
ching,
‘ypical
ntified
future
ligital
new
ears to
highly
atured
digital
7 On a
he use
tcopy
)pular
s well
Image matching can be classified into different techniques
depending upon the features used to detect similarities. The
most popular is area-based matching in which a gray value
matrix of one image is compared to a gray value window of
another image on a pixel by pixel basis. This method is most
accurate for measuring image coordinates of well defined
points as it gives sub-pixel accuracy. However, this method
also requires very good approximations (about 5 pixels) in
order to find matches successfully. Other matching techniques
are based on features which could be lines or characteristic
points of the image. They must first be extracted before the
matching can begin. This technique is less accurate, but more
robust since correct matches can be found over the whole
image area without any approximations.
Other types of image matching, such as binary tree matching,
are currently under development and have not been practically
implemented in digital photogrammetric systems. The
matching of points can take place in image space (on the
sensor) or in object space (on the map or on the ground). The
relationship between image and object is given by the
collinearity (perspective) equations, which form the basis of
all photogrammetric point positioning methods. Any point-
pair matched in image space can be immediately projected to
the ground.
Once in object space, the 3-dimensional points can be used to
generate a raster DEM stored in the same format as digital
image (a matrix of gray values). Photogrammetry has been
involved in the development of DEM interpolation techniques
for a long time. Good DEM interpolation techniques allow us
to approximate the terrain smooth functions by using reference
points. It is very important to smooth reference points derived
automatically in order to eliminate wrong matches and reduce
noise. Analytical surfaces developed during the DEM
interpolation process are represented as a dense grid of
elevation points in the GIS. At each location of an elevation
pixel, the corresponding gray-value can be found by digital
orthophotography. The surface point represented by the DEM
pixel is projected into the original image using the perspective
Ground Camera
Control Parameters
relationship between image and ground. The location in the
original image is identified and its corresponding gray value is
extracted by a resampling technique. Once completed over the
whole image, the digital orthophoto is created. This layer
fully corresponds to a (digital) map free of any relief
displacements.
Digital photogrammetry is heavily involved in the development
of feature extraction and image understanding techniques. The
digital mapping of linear features such as roads, rivers or
rectangular objects such as buildings is of profound
importance. Research is currently underway to automatically
find these lines, which are usually highly visible in images and
supposedly correspond to edges of the real surface. Once
lines have been detected and vectorized, artificial intelligence
has to be applied to automatically interpret their meaning.
Image understanding is a popular discipline which will soon
allow us to fully and automatically analyze vector data
obtained from the feature extraction. This data can then be
integrated into the GIS together with the raster information.
IMPLEMENTATION OF A DIGITAL
PHOTOGRAMMETRIC MODULE IN A RASTER
GIS
Many of the techniques described in the previous section were
implemented in an existing raster GIS to allow the user to
acquire spatial data directly in the familiar GIS environment
using digitized aerial photographs or satellite imagery. They
are collected in a digital photogrammetric module, mainly
applied for information extraction for the GIS data base. The
functions are comprised of coordinate measurement in the
imagery, control data collection, sensor orientation for aerial
photos and satellite imagery, image matching, DEM
interpolation and orthophotography. The user can densify
control point networks by aerial triangulation or satellite
triangulation, create a digital elevation model by image
matching and DEM interpolation, and consequently derive a
digital orthophoto by using the DEM to correct for relief
displacements. Figure 1 below shows a flow chart describing
all of the functions implemented in this module.
Image Scanner
Coordinates Calibration
| > nt iles
Aerial or Satellite
Triangulation
|
Image
Matching
DEM
Interpolation
|
Orthophoto
Generation
Orientation Parameters
Random DEM (Reference Points)
Raster DEM
Digital Orthophoto
Figure 1: This diagram shows the digital photogrammetric functions integrated in a raster
GIS. The boxes indicate the functions and the text to the right state to the data derived at
each stage.
895
