There are two measurement modes: one is manual
and the other is an automatic DTM (Ohtani, Ishii,
1992). The latter uses the stereo-matching by
coarse-to-fine cross-correlation. This is a mode
wherein a computer makes batch measurements by
placing evenly spaced grid points on the left
image to be processed by correlation.
Out of the calculated 3D coordinates we can
obtain the stereo-contour-lines, perspective view
and cross-section of the object.
Furthermore, since the measurement data can be
output by the DXF format, our system a whole
variety of applicability through AutoCAD*,
CivilCAD(TOPCON), and the CAD system which
is supporting DXF*.
2.3 Camera Calibration
For camera calibration we used the
self-calibrating bundle adjustment software, which
we had already developed. We have also developed
a new software to detect automatically with
sub-pixel precision the position of the targets
placed on 3D test-field designed for calibration.
This software establishes the correspondence
between the image coordinates and the 3D
coordinates of the targets already measured on 3D
test-field by means of the orientation of single
photograph. The software also detects the coarse
position of the targets by means of image
correlation as well as its fine position by
detecting the edge with Laplacian-Gaussian filter.
This 3D test-field (Fig.4) was 1000 mm in both
height and width, on which were placed 121
targets "in 3 depth stages, i.e., Omm, 100mm,
200mm, 300mm, 400mm and digitally
photographed from 7 different positions
(Cf.Fig.5). We used the 3D coordinatesc of |: the
targets measured by a contact-type 3D
measurement apparatus at the precision of = 10u
m.
Now, as to the details of camera calibration,
confer to (Kochi, Ohtani, Nakamura and others
1995).
Fig.4 3D test-field
*AutoCAD,DXF are U.S. registrated trade marks
of Autodesk, Inc.
78
3D test-field Camera position
Fig.5 Configuration of image acquisition
3. SURFACE FLATNESS MEASUREMENT
Lately, the exterior characteristics of train body
are increasingly diversified. For example, often
times its surface is plain without design, or sandy
with sandblasting or simply lustrous, or three
dimensionally deep with curve, many of which
characteristics make it difficult to measure by
automatic stereo-matching.
Therefore, we first executed a preliminary testing
with various simulated surfaces to see whether our
surface measurement system really works and then
applied it actually on a real train body.
3.1 Simulated Surface
We had already measured 80 points on a simulated
surface of plain iron plate of 400mm X 300mm
with the base length of 400mm and with the
distance of 1000mm and obtained a satisfactory
result of depth accuracy 0.39mm(rms) and target
accuracy (Cf:(1)equation) 0.48mm (Kochi, Ohtani,
Nakamura and others 1995).
This time we went further as to test-measure the
sandblasted lustrous simulated surface of 300mm
X 300mm with a round protrusion in the center to
assess the precision (Cf:Fig.6).
On a sandy surface ordinarily the light from its
source is diffused and reflected on the different
parts of stereo images of the right and left
cameras, thus causing the difference in ‘the
shading between the obtained images of two
cameras. However, in our experiment it did not
affect our stereo-matching much to . our
satisfaction.
The Fig.7 shows the cross section obtained from
the measurement results processed through
PI-1000. To confirm the accuracy of our system
this result was compared with the result obtained
by the contact-type-3D-measurement-apparatus of
- $ y m on 80 different points on the same
surface. In our measurement the base length was
928mm with the distance of 1048mm. The
experiment was quite satisfactory with the
accuracy result of 0.23mm (rms) in depth
measurement, compared with the targeted
accuracy of 0.23mm.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
The 41:
followi
8 Z-(E
Ô Z:tar
(d
Ô p:acc
readi
COOI(
H:dista
came
B:base
f :princ
3.2 Ap
We ex]
body a!
Co.Ltd
800mm
conditi
the act
perspe«
measur
As the
strongl
thus fo
to 60(
excepti
satisfa:
was 0.!
of 0.1n
targete
