ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
94 
2.1 GENERAION OF NEW DSM AND OLD DSM 
DSM is a digital representation of the surface of the earth. The 
difference between DSM and DEM lies in that the elevation of 
DSM contains the height of objects of earth surface but DEM 
does not. DSM can be quickly and conveniently generated on a 
digital photogrammetry workstation. In this experiment, we 
utilized VirtuoZO Digital photogrammetry workstation to generate 
DSM for the new and old stereo image pairs of the same urban 
area. The detained steps to generate DSM on Virtuozo was 
referred to references (Zhang , 1994). Figure 1 shows the 
parallax contours from partial result of the image matching, 
where buildings appear as lumps. Old and new DSMs are 
generated based on the result of the image matching. 
Figure 1 the parallax contours from partial result of the image 
matching 
2.2 ACQUISITION OF CANDIDATE CHANGED REGIONS 
The two DSMs are subtracted to generate a difference of DSMs. 
Contour lines can be generated from the difference DSM. In the 
contour graph, groups of plenty of nested densely contour circles 
appeared as "lumps", which represented an obviously changed 
regions. “Lump” boundary’s extraction could be done using 
traditional region tracing algorithm. According to the collinearity 
equations and the known interior parameter of orientation and 
exterior orientation, the potentially changed image regions could 
be rapidly projected on original images. Figure 2 shows the 
“Lumps” arising in the regions with height eminently changed. 
Figure2 “Lumps” arising in difference DSM 
2.3 EXTENSION OF CANDIDATE CHANGED REGIONS 
Because of the complexity and difficult of the urban images, the 
candidate changed regions, deriving from the image matching, 
may not contain the whole changed building sometimes. They 
should be extended to be suitable for the next processing. 
Region growing algorithm was used for the extension (Zhang, 
1996). The thresholds determined by the maximum between 
class variance analysis were used to guide the extension 
process.. 
(b) 
Figure 3b shows the extended regions from those shown in 
Figure 3a. 
Figure 3 Extension result of the candidate changed area