ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
of experiment and verification, some basic rules and threshold 
can be determined by experience. For example, a simple rule 
could be: if this difference is distributed in more than 3 direction, 
their frequency > F min „ change of building must took place in this 
region. After these rules were decided, It was easy to detect the 
existence and change of buildings for every region. Due to 
limited space, the rules we adopted no longer listed here in 
detail. 
ft was simpler and easier to compare than the old one. 
1 73: 12 1 
7 : 103 
83 : 63 
263:57 
90 180 270 360 
(a) 35# region (new period) 
7:32 
83:39 
90 
263:12 
180 270 
360 
(b) 35# region (old period) 
Figure 6 four gradient direction histogram 
3. EXPERIMENTAL RESULTS 
we experimented methods presented in this paper on two stereo 
photo pairs of two different period of one urban area. All together 
151 regions were extracted from the difference of DSMs. After 
extending those candidate regions, some intersected regions 
were combined and some very small regions were removed, 
altogether 97 regions was regained for afterwards analysis. 
Among 97 regions, for building, 41 regions are really changed, 
56 regions really not changed, the gradient direction histogram 
was 
(a) epipolar Image of new period 
(b) epipolar image of old period 
Figure 7. Result of Change Detection 
analyzed to determine really changed regions and change’s type. 
Experiment result showed 31 changed regions to be detected 
correctly and 41 not changed regions to be detected corrected. 
The absolute accuracy of detecting according to formula 
suggested by David (David, 1981 ) reached to (31+41)/97 = 73%. 
Figure 7 shows a part of the result. The regions with white 
boundary in top image window of figure 8 are changed, and the 
regions with gray boundary are not changed. 
4. CONCLUSIONS 
The above analysis and processing were executed automatically 
and did not need users’ intervention. Experiment’s method was 
simple, easy to operate and meanwhile result of the experiment 
was also encouraging. 
Because of making use of both techniques of stereo image 
analysis and single image analysis, the method presented in this 
paper was more reliable and had a better performance than the 
single image analysis. Of course the detection ratio is still low. To 
raise the correct ratio of detection, there are a lot of problems to 
solve. 
ACKNOWLEDGEMENTS 
Thanks for the supporting from Natural Science Fund of 
P R.China (No. 49771063). 
References: 
[1] Singh A., 1989, Digital change detection techniques using 
remote sensing, Int, Remote Sensing, Vol. 10, No.6, 989-1003 
[2] Fung, T, And E. Ledrew,1987, Application of Principal 
components Analysis to Change Detection. Photogrammetric 
Engineering and Remote Sensing, Vol.53, No.12, p1649-1658 
[3] Cushnie, Q., 1989, Monitoring urban land cover changes at 
the urban fringe from SPOT HRV imagery in south-east England, 
Int. Remote Sensing, Vol.10, No.6, 953-963.