International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 . 
  
However, actually, noises such as electric poles, electric 
cables and trees exist in textures in many cases, even if 
nadir images are used. Therefore, the threshold value of 
change detection should be set a lower value. In our 
preliminary study, a best threshold of building change 
detection is 0.8-0.9. Based on this processing, building 
demolition and reconstruction can be detected. The 
correlation coefficient is less value than the threshold, 
when buildings are disappeared or constructed (See, 
Figure.8). However, a case, which a new building is built 
up in a vacant lot, cannot be applied to this change 
detection algorithm, because initial polygons do not exist. 
In this case, lot boundaries obtained from GIS data can 
be used for this processing. 
Existing image 
Latest image 
     
Projection 
Back- 
  
  
  
  
  
Correlation coefficient = 0.5370 
(R=0.523 G=0.579 B=0.509) 
Figure. 8. Example of preliminary 
study (Comparison of roof polygon) 
6. NEW 3D DATA CONSTRUCTION 
While buildings with changes are detected in the existing 
TLS images, new 3D data are reconstructed at the same 
location by using initial values. The initial values are 
based on DSM generated from TLS images, a building 
template model and surrounding information. Based on 
this algorithm, a feature based matching is applied to 
reconstruct new buildings. However, especially in urban 
dense area, it is difficult to detect building boundaries 
due to a complexness of a building feature. Therefore, 
authors have developed the semi automatic matching 
based building extraction application. This application 
requirès only polygon input manually in a single image 
without stereoscopic measurement, and it generates 3D 
data automatically, as Figure.9 shows [3]. 
  
  
     
  
e se 
  
Rz ry 
Figure. 9. Semi-automatic matching based building 
extraction application 
7. CONCLUSION 
The method of revision 3D building data by integrating 
texture change (roofs and walls) and 3D shape change of 
buildings using STARIMAGER/TLS (Three Line 
Sensor) is proposed in this paper. When high-level 3D 
data are prepared beforehand, this method is effective for 
automatic change extraction of 3D building data in urban 
dense areas. 
8. ACKNOWLEDGEMENT 
STARIMAGER / TLS images were provided by 
STARLABO Co. Ltd. The authors thank this company 
for acquiring these data available. 
References 
[1] Masafumi NAKAGAWA, Ryoruke SHIBASAKI, Development of 
Methodology for Refining Coarse 3D Urban Data Using TLS 
Imagery, ISPRS Commission | |, WG 6, 2003 
[2] Masafumi NAKAGAWA, Ryosuke SHIBASAKI, Y.KAGAWA, 
Fusing stereo linear CCD image and laser range data for building 
3D urban model, ISPRS Commission _, WG _/7, 2002. 
[3] Katsuyuki NAKAMURA, Masafumi NAKAGAWA, Ryosuke 
SHIBASAKI, 3D Urban Mapping Based On The Image 
Segmentation Using TLS Data, 23™ Asian Conference on Remote 
Sensing, 2002 
[4] M.NAKAGAWA, H. ZHAO, R.SHIBASAKI, Comparative study 
on model fitting methods for object extraction, Asian Conference 
on Remote Sensing, 2000