ustry 
camera 
nearity 
e lenses 
fidelity 
nents or 
nmetric 
of the 
ts have 
us the 
rocess, 
'hich is 
)mpass 
sed for 
At the 
xterior 
1 mm 
> been 
em has 
b. The 
ly and 
The 
with 
itution 
vs that 
lustrial 
Its of 
  
2.1 The stereo-correlation technique 
Binocular stereovision is a technique for building a three 
dimensional description of a scene observed from two slightly 
different viewpoints. From a pair of images, it is possible to 
compute the 3D coordinates of a physical 3D point by 
triangulation under photogrammetric process. 
An important task in 3D computer vision is camera calibration, 
especially when metric information is required for applications 
involving accurate dimensional measurements. The proposed 
technique only requires that the camera observes a (quasi 
planar) pattern shown at different orientations. The motion of 
the pattern need not be known and the pattern itself can be 
imprecise. This technique has been recently extended to the 
calibration of a stereovision sensor. Using a photogrammetry 
approach, the intrinsic parameters of each camera, the 3D points 
of the pattern and the relative position and orientation of the two 
cameras are computed all together using bundle adjustment. 
The calibration parameters will be used in the sequel at different 
stages: 
e the correction of lens distortion, 
e the calculation of the 3D position of a scene point 
from its stereo image by least square image matching. 
2.2 Rectification of the pair of stereo images 
This is the stereo-matching problem, it appears that second 
image points which belongs necessary to a straight line of first 
image, entirely defined by the coordinates of first image point 
and the relative geometry of the two cameras, called the 
epipolar line associated to point on first image and second 
image also [Ayache,1988 ][Loop,1999 ]. This geometric 
constraint imposed by the imaging system, called epipolar 
constraint, the epipolar lines form a bundle of lines going 
through an epipolar center of first image which is the image of 
the optical center in camera second. Camera calibration 
algorithm is geometric calibration method which have used for 
control points and self calibration adjustment with additional 
parameters within the test field at Photogrammetry Department 
in LT.U. Consequently the distortion parameters have been 
calculated with block adjustment until the correlation 
coefficient between the inner orientations parameters are zero. 
In first exterior orientation for the first stereo image pair, some 
signal points have been measured with compass (0.02 mm 
precision). These edges have been used for the restriction in the 
bundle adjustment process. At the end of the bundle 
adjustment, the results for the coordinates of projection centre 
and rotations have been calculated accurately. In the particular 
case where the image planes are coplanar and parallel to the 
vector of centre of the images defined by the optical centers, 
then the epipolar centers are rejected to infinity and the epipolar 
lines form a pencil of parallel lines. For camera station 
configuration, the epipolar lines have been parallel to the axes 
of the image coordinate frames. This ideal configuration 
enables ^ efficient stereo-matching procedures since 
corresponding pixels are on the same row in both cameras. In 
practice, it is impossible to get the ideal configuration of the 
cameras mechanically. 
Nevertheless, the other method is possible to apply to each 
image of the initial stereo pair a transformation, called 
rectification, to obtain a new pair of stereo images 
corresponding to a virtual stereo rig with perfectly aligned 
cameras [Loop, 1999]. The rectification procedure uses the 
calibration parameters computed in the offline camera 
calibration phase. 
     
- | + . \ 
Figure 2: Slab Surface and CDP’s 
When reinforced concrete slab, data capturing have been started 
and the LVDT have been used for the displacements on the slab 
surface have been measured simultaneously. The 
videogrammetric data have been captured same time period 
with the force which applied to thrthe slab. 
The photogrammetric process has been started after the 
experiment. For each image pairs have been oriented with the 
same exterior orientation parameters which had been calculated 
from first image pairs. 
3. COMPARISION OF THE RESULTS 
TML CDP 50 transducers have been used for the eight different 
points on the slab surface for determining the differences. TML 
CDP 50 (Linear Variable Differential Transducer) have 
determined the X,Y,Z displacements with 0.005 mm precision. 
The principle point coordinates and its accuracy for bundle 
adjustment from PMM (Developed software by 
Photogrammetry Department of I.T.U. ) have been tested with 
Pictran and Australis Software. After this checking, the 
resection results from PMM (Photogrammetric Measurement 
Machine) have been compared with the results of the LVDT. 
However, the compared coordinates have not been the same 
signal points; these signal points could not have been measured 
by photogrammetric method so that LVDT covered these 
points. Hence, genera surface deformations have been obtained 
with LVDT and photogrammetric method. Also the DTM of 
the surface have been obtained from these data. The software 
PMM (Photogrammetric Measurement Machine) have been 
developed by I.T.U. Photogrammetry Department for making 
industrial measurements in the expert system which has been 
build in Photogrammetry Department. This expert system has 
been developed for applications of the very close range 
industrial practises. 
Measurement Table 
+ Photogrammetric 
LVDT 
    
Force [kN] 
0 5 10 15 20 25 30 
Displacements [mm] 
Table 1: Measured Displacements 
—19—