1ge
om
vil
om
the
1ge
rid
rid
the
on.
3D
95
257 .
3.3 Experiment Results
The interior orientation of the camera whereby the experiment results were obtained is illustrated in Table 2.
The lens distortion is that of barrel distortion.
Also, the 3D accuracy obtained by interior orientation data is as in Table 3, based on the pair of images taken
from the positions 1 and 3 of Fig.7. Here, dxy denotes the horizontal direction, and dz denotes the depth.
Table 2. Interior orientation parameters
Further, the target accuracy was calculated by the
following equations. Principal distance Sf 36.68518mm
Radial distortion aberration kd -7.1732E-5
&XY-Z8p, 6Z-Ze.P8p (5) 12 5.9775E-8
Tangential distortion aberration : pl -8.1542E-6
> D2 -1.2556E-5
8p : accuracy of reading the image coordinates
H :distance between camera and object — B : base length Table 3. 3D measurement accuracy
From these results, it has been confirmed that our 3D 9 X Y (rms) | Ó Z (rms)
measurement apparatus has achieved remarkable |Target accuracy 0.161mm 0.298mm
accuracy of 0.3mm/m for a digital system. This CCD
camera showed itself capable of camera calibration in the — =
equal perfection as that of a film camera. With inner orientation correction |0.072mm 0.190mm
No inner orientation correction 0.557mm 0.0834mm
4. EXAMPLE OF 3D MEASUREMENT
4.1 Overview
Based on the aforementioned calibration and measurement accuracy results, a study was conducted to see whether
it would be possible to measure warping (plane irregularity) of the body of an automobile, for example, and what
sort of accuracy would be obtained. For this experiment, a plain iron plate of approximately 400x 300mm in
dimensions was purposely warped to serve as a part of an automobile body, and the measurement was conducted
thereon. In order to verify the accuracy, this iron plate was attached to the 3D test-field of Fig.6, and measured
with a contact-type 3D measurement apparatus, beginning from one edge, following a grid pattern of 40mm each, at
a total of 80 points. Stereo image was taken by high-resolution CCD camera at a base length of 40cm and at a
distance of Im from the camera. The stereo image was analyzed by PI-1000 and the accuracy comparison was
conducted. Exterior orientation was conducted by employing the Table 2 calibration data, and as to the control
points, we used 10 points which had been left out of the iron plate on the 3D test-field. As it is impossible to make
a comparison without some kind of design on the iron plate, a certain design was projected on the iron plate by a
projector, so that stereo-matching and measurement could be conducted.
4.2 Measurement Results Table 4. 3D measurement accuracy
Sxy {rms) 182 (rms)
Target accuracy 0. 2mm 0. 48mm
For measurement, after approximately 14,000 points
were subjected to stereo matching with the automatic DTM
measurement mode of the PI-1000, the mismatched points |Measurement results | O. 0 0 3mm 0. 39mm
were manually corrected. Table 4 illustrates the accuracy
obtained in 3D measurement. As can be seen from the | : |
Table, the results fully met the target. | oe
Further, Fig.8 illustrates an error histogram, where the ee PTT
contact-type 3D measurement apparatus is giving the true |
value. The areas where the accuracy was low were mostly |
towards the edge of the iron plate. This was due to the fact ||
that the points outside of the plate itself were also taken in |
and thereby causing the mismatching. There is also the |
possibility of interpolation error within the PI-1000. |
However, either of these errors is easy to avoid, and further |
improvement in accuracy is possible. |
SE SS
0.4 0.6 or more
0,5. 70.6 — -0.4
-0.2 0.2
residual (mm)
Fig. 8 Error histogram
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995
