Table 5 Time required for revision work — Non-experienced
Aerial photo image Area A | Area B | Area C | Area D
with digital map overlay
(no other pre-processing)
with masks
70 90 120 70
60 80 120 60
with masks and emboss
filtering 40 70 90 45
unit: minute
Table 6 Time required for revision work — Experienced
Aerial photo image Area A | Area B | Area C | Area D
with digital map overlay
(no other pre-processing) | 30 35 54 30
with masks
31 30 55 32
with masks and emboss
unit; minute
The houses to be added in the revised map were checked by
stereoscopic observation of aerial photos and the number of
them in each test area is shown in Table 7.
Table 7 Number of houses to be added
Area A Area B Area C Area D
144 73 258 141
To see the accuracy of each method, number of omissions and
erroneous inputs are summarized in Table 8-11 for each
operator.
Table 8 Number of omissions — Non-experienced operator
Aerial photo image Area A | Area B | AreaC | AreaD
with digital map overlay
(no other pre-processing) | 2 13 2 4
with masks
6 9 3 1
with masks and emboss
6 15 5 0
filtering
Table 9 Number of omissions — Experienced operator
Aerial photo image Area A | Area B | Area C | Area D
with digital map overlay
(no other pre-processing) 0 5 0 0
with masks
1 15 0 0
with masks and emboss
filtering 0 9 1 0
Table 10 Number of erroneous inputs — Non-experienced
Aerial photo image Area A | Area B | Area C | Area D
with digital map overlay
(no other pre-processing) 1 4 6 1
with masks
0 8 1 2
with masks and emboss
filtering 1 11 7 2
Table 11 Number of erroneous inputs — Experienced
Aerial photo image Area À | Area B | Area C | Area D
with digital map overlay
(no other pre-processing) 1 2 4 1
with masks
0 4 2 1
with masks and emboss
filtering 1 17 9 2
The effectiveness of masking is not clear from the result (Table
5 and 6). Some speeding up was observed for a non-
experienced operator but no effect was observed for an
experienced operator. One of the problem of masking is that the
number of omissions increased by the experienced operator in
area B where many houses already existed. It was found that
about 30 % of these omission was caused by being buried under
expanded masks. Therefore the amount of expansion must be
carefully chosen.
Emboss filtering has considerable effect of speeding up for both
non-experienced and experienced operators. But the number of
erroneous inputs increased. It is therefore hard to justify simply
adopting the emboss filtering.
4.3 Discussion
In this method, human operators digitize the line of houses
looking at CRT screen. It was felt by the operator that reference
photo images should have higher resolution than 0.5 m on the
ground. Otherwise the outline to be digitized is unclear and
emboss filtering also do not generate clearly visible outline.
Emboss filter is effective for finding houses constructed newly
on bare lands. But it is not much effective for finding houses in
already developed area. Some other method should be
developed for the case.
The experiment was for detection of new construction only. To
apply inverse of the above masking image can be used to detect
disappearance of houses.
5. CONCLUSION
Several methods have been tested to detect changes from image
processing of aerial photos. Some of them are promising in
view of reducing human work load. It seems that height data
obtained by automatic stereo matching give useful information
on land use change. Study in this direction should be further
pursued aiming at adaptation to practical mapping process.
Image understanding of aerial photo is very difficult problem
and it is impossible to get practical result only with this
approach. Therefore it is important to combine every available
data source in order to get practically useful result. This multi-
data fusion approach, in particular effective utilization of
existing digital cartographic data, should be the subject of the
next phase.
REFERENCES
Geographical Survey Institute, 1996. Study on semi-automatic
analysis of aerial photos for land use change detection (The rii
year). Geographical Survey Institute, Tsukuba. (in Japanese)
Geographical Survey Institute, 1995. Study on semi-automatic
“analysis of aerial photos for land use change detection (The 1%
year). Geographical Survey Institute, Tsukuba. (in Japanese)
Oyama, Y., 1996. Semi-automatic digital photogrammetric
'system on PC. In: International Archives of Photogrammetry
552
and Remote Sensing, Vienna, Austria, Vol.XXXI (to be
published).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
KE
Ab:
pho
COO!
pho
aerc
acqi
GIS
its €
top
mea
visu
of tr
that
data
So f
whi
phot
P ho
nece
phot
mod
pape
capa
PHC
inte,
