for both photos, so the combined horizontal 
displacement measurement error was estimated 8cm. 
The vertical error was 8cm for the pre-photos and 
14cm for the post-photos. The latter large error 
was caused by the incomplete models produced along 
the seashore. Without these, the combined vertical 
displacement error was estimated to come within 
15cm. 
4.3 Scanning photos 
For the experiment two sheets of pre-photos and 
three post-photos were scanned with two scanners. 
One is VEXCEL VX3000PLUS with the nominal 
precision of 3um and the other is ZEISS PhotoScan 
(by courtesy of Geographical Survey Institute) 
with the precision of 2um. These photographs cover 
a common area of about 500m*800m. 
Since the required measurement precision was 10cm 
or higher, 10 um pixel size ( 4-5cm on the ground) 
was thought necessary. It produces image data of 
about 600MB/film. Though the scanning time was 
only 30min, it took 4-5 hours/film to write out 
the data to DAT(Digital Audio Tape). Eventually 
with VX3000PLUS the images were scanned with 10um 
pixel size, but with PhotoScan, 15um pixel was 
used because of the limit of exclusive use. In 
this case the volume was reduced to only 
200MB/film. 
The second difficulty was a small dynamic range of 
CCD sensors. Fiducial marks were thinner imaged. 
When the sensitivity was adjusted to fiducial 
marks, details of scenes become hardly recognized 
and looked nearly binarized. In this experiment, 
the image quality of fiducial marks were 
sacrificed to the extent of the border. But the 
densitometric range of obtained digital images was 
still short. 
4.4 Orientation and Measurement 
The practical measurement was done with an 
analytical plotter WILD BC-1. GOKUU could not 
attain the precision of BC-1 in inner, relative 
nor absolute orientation. The errors with GOKUU 
in interior orientation were 1/5-1/4 pixels and 
the residual y-parallaxes in relative orientation 
were 1/3-1/2 pixels by the MLSC for the three 
models, while with BC-1 the interior and relative 
orientation errors were both 2-3 um. 
Errors in absolute orientation varied largely with 
image quality. The orientation points (GCPs and 
supplementary GCPs) had been marked on one side of 
a pair of photographs by a pricking device. But 
50% of points were invisible at orientation with 
GOKUU, because these points had been taken on 
poorly textured flat places, e.g., on bare soil or 
paved roads. Though these pricked holes could be 
observed stereo-optically clearly with BC-1, it 
150 
was hardly seen on the display. This implies that 
not only the dynamic range of CCD sensors is 
short, but the quantization level of 1 byte is 
also insufficient. The absolute orientation with 
GOKUU gave the 8-9cm( rms) both 
horizontally and vertically, while with BC-1 the 
residuals were as half as or less with GOKUU as 
shown in Table3. The effect of pixel 
concealed by the variance of 
orientation points. 
error of 
size was 
image quality of 
For keeping the 
degradation we 
orientation precision from 
took again supplementary GCPs 
which were clearly identifiable on the display for 
the three models, and then each set of the 
photographs were adjusted at once by the bundle 
adjustment including these points. The correspond- 
ing points identified for displacement measurement 
were also included as supplementary GCPs. 
To identify corresponding points in the pre- and 
post-photos with GOKUU, two operators observed the 
images on two displays cooperatively. These points 
include corners of lines of pedestrian crossings, 
stop lines, letters (STOP, 50KM etc.) and corners 
of grid lines in parking lots. Since there are 
three years' gap between two flights and the 
flight direction differ by 90 degrees, careful 
observation was necessary. Lines or letters might 
be often rewritten. Roads might be repaved. But 
most of them were easily discernible by changes in 
tone or shapes of lines. 
The same points were also measured with BC-1. Fig. 
6 shows the differences of observed ground 
coordinates between the two plotters for one pair 
of the pre-photos. The differences are accredited 
point measurement errors coupled with orientation 
images. The difference of horizontal 
errors is estimated to reflect the pixel size. The 
errors of 
points with a more than 30cm error are probably 
mis-identified. These figures show GOKUU can not 
compete with BC-1 at present. 
5. MEASUREMENT OF THE GROUND DISPLACEMENT IN 
ASHIYA-CITY with BC-1 
In November, 1996 we determined to change the plan 
of all the measurement with GOKUU to that with 
BC-1. About 10-15 supplementary GCPs were includ- 
ed in each model. To avoid mis-identification of 
points between a pre-photo and a post-photo, we 
looked for the identifiable corresponding points 
(or their candidates) on double enlarged prints of 
the pre- and post-photos. And all the points were 
sketched. 
About 700 points in totality were measured. The 
points were taken near the lattice of 100-200 m 
width. This measurement work took for two opera- 
tors about three months. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
Thou 
coul 
inte 
root 
manh 
plac 
thos 
Sinc 
is i 
sign 
case 
as | 
tren 
afra 
and 
vert 
This 
tion: 
a sp: 
Corre 
GOKUL 
the 
meast 
eartl 
with 
expel 
reso’ 
prec] 
prove 
compe 
iden! 
taker 
Acker 
Photc 
pp.42z 
Hattc 
auton 
Conti: 
Hattc 
ment 
Basec 
Close 
Melbc 
Ianni 
Gener 
Operz 
Nolet 
DVP: 
58(1)