2. IMAGE DATA ACQUISITION
In order to make use of the advantages of digital photo-
grammetry it is necessary to acquire the image data
directly from the SEM without any non-defined signal pre-
processing. This fact is of importance because of the high
requirements in accuracy of 3D-measurement. Beside
this, a direct connection enables the operator to choose a
suited area for photogrammetric surface determination
and offers an image acquisition with optimal conditions
concerning to image quality.
A digital interface connecting a personal computer (PC)
and SEM was developed in a joint project between the
Technical University of Berlin (TUB) and the Institute for
Physical High Technology (IPHT) Jena (Fig. 1). The con-
nection consists of a AD/DA-converter for raster control
and image data acquisition, and a serial interface for con-
trolling and saving pickup parameters (for instance wor-
king distance or magnification). Therewith it is possible to
take tilting series of images — necessary for photogram-
metric 3D-evaluation — directly from a PC.
This interface offers also new potentials for the improve-
ment of image quality. By controlling the scan of the
electron beam in dependence on the signal level (which
means the grey value in image data) the images have a
higher contrast — even in dark and shading areas of the
scanned microprobe.
3. PHOTOGRAMMETRIC PROCESSING
The photogrammetric evaluation requires two or more
SEM-pictures of a probe under different tilting angles.
Photogrammetric processing takes place successively in
different steps. Besides this the selection of the appro-
priate approach for an automated 3D-determination of the
surface depends — as already mentioned — on the charac-
teristics of the microprobe.
Electron Beam
Microstructure with
+10° tilt angle
Microstructure with
-10? tilt angle
Image 1 Image 2
Figure 2: Principle of Parallel Imaging with the SEM
226
3.1 Parallel-Block-Adjustment
The basis of the photogrammetric processing is the esti-
mation of the orientation parameters. In general a mathe-
matical description is used, which allows six degrees of
freedom for every image: three rotations and three trans-
lations. Because of the applied geometrical model, which
is a parallel projection, the number of degrees of freedom
is reduced to five. Additionally there are parameters for
image distortion and scale available (different in x and y
direction), because the conditions of image acquisition do
not remain constant in the SEM.
The block adjustment requires homologue image points in
all pictures involved. According to the used microprobe it
is necessary to select the appropriate procedure for this
task. The easiest method is to measure the required
points interactively in each image. Correlation techniques
simplify this procedure by determining and measuring
image points in corresponding images with subpixel ac-
curacy. In order to achieve a higher degree of automation
in photogrammetric processing it is possible to use fea-
ture or edge extraction operators in relation with feature
matching methods. The results of the automatic measure-
ment of homologue image points will be presented below.
Because of the changing imaging conditions of the SEM it
is necessary to determine the scale factor for each image.
Additionally, imaging with the SEM allows no definition of
control points. Hence, the parallel-block adjustment takes
place in form of a self calibration. In order to enable self
calibration it was necessary to use least-squares methods
for free bundle adjustment. In this case the spatial distri-
bution of the tie points plays an important role for the
solution and accuracy of the block adjustment.
The orientation process runs in two steps. The first step is
the estimation of approximate values. Given are the cor-
responding image coordinates of the tie points and ap-
proximate values for the tilting angle and the scale factor.
On the assumption that the images are only tilted, it is
possible to determine approximate coordinates for the
used points in object space (Burkhardt, 1981):
Xi 7 Xi
Xi Yi sinß
m, m m
p
(1), (2), (9)
X
X,y,2 object coordinates
XV, image coordinates in image i
X,,,y4. image coordinates in image i--1
tilt angle of image i
m, m, scale factor in x- and y-direction
Following, it is possible to define a coordinate system in
object space, necessary for instance for the definition of a
reference plane for a Digital Surface Model (DSM).
The second step is the determination of the orientation
parameters and the scale factor for each image. The
orientation process is performed in analogy to central
projection in form of a parallel block adjustment, because
of the parallel imaging equations:
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
Xj =m, (bi
y "m, : (bz
Xoyo . COO
b,-b, elem
In order to c
crease the s
sible to meas
to use it in the
In addition tc
affine factor
through this |
al., 1994). Us
perform a ca
performed b
probe. The r
lowing steps
constant ima
scaling facto!
3.2 Area-bas
Homologue i
continuous s
sured with ar
|
|
|
The applied
yielding reli:
texture (Kor
method in tw
a) Normalise
b) Least-Squ
In case of
matrix is shi
a correspor
efficient is c.
coefficient
homologue |
The results
the approxir
