The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
392 
Figure 1. Picture of airship 
In order to fly more smoothly and reposefully, unmanned 
airship should be controlled by pre-programmed automatic 
controller when it is flying in the air. And most important, for 
the sake of auto-aerial triangulation (AAT for short) which is 
the key to efficiency and real time process and also the key step 
to this thesis, it is commanded that photos taken from airship 
should have about 80% degrees of forward overlap and 60% 
degrees of side overlap. As figure 2(A,B) shows. In this case, 
there are enough corresponding points and tie points between 
adjacent photos and stereo models. 
Base Line 
There is something else we should pay attention to: Some 
building changes may not great. They may have height changes 
of one meter or 3 meters. So when design the flying height, the 
precision must be considered in advance. 
4. INDOOR WORK 
When flight is finished, with Airborne Image and POS data was 
recorded, it is time to get read for auto-aerial triangulation. First, 
check data: Is there any data lost? Is any error exits? Is the 
quality of the data meets the requirements? They all should be 
examined carefully. If everything is ready, or the error is filtered 
out or correct, then gather the data into project, start up AAT, 
leave the work to the computer. According to our own software 
MAP-AT, and hardware: windows XP system, 1G memory, 
1.8G inter CPU, the processing shows that: about 120 stereo 
images can be proceeded in a hour with image size 3000 pixels 
multiply 4000 pixels. 
Then, raw DSM can be produced though epipolar images and 
geometrically constrained feature point hierarchical Matching. 
Next, after process of relaxation matching, the final high 
precision DSM can be obtained. 
Figure 3. Work flow of building change detection] 
At last, what is left is to get building changes in different period, 
and what should be done is to get DSMs from different period 
and make comparison. For example, if you have DSM produced 
two weeks ago, and now you get another DSM today, subtract 
one from another, then building changes are marked with 
different colour, and different colour shows different height 
change: Green colour means new building with height change 
greater than 1 meter. Red colour means tom down building with 
height change greater than 1 meter. Blue colour means new 
building with height change greater then 5 meters, and so on. 
And the colour can be customized according to user’s custom. 
Then we get the work flow of field and indoor word (all of these 
constitute the flow of building change detection). The flow 
can be simply showed as figure 3. 
5. EXPERIMENT 
In our experiment, Height changes of greater then 3 meters are 
set to be detected. And the results of experiment on test city 
show that: what was proposed in advance is feasible and 
practical. All building height changes which are greater than 3 
meters and less then 5 meters are detected. The hardware and