Figure 2: stereopair including box building and hip-roof 
building.( taking from the URL:www.middlebury.edu/stereo.) 
Figure 5: project the dsm data to the left (or rightjimage edge 
map to refine the edge fo the coarse model 
6. CONCLUSIONS 
f 
Figure 3: disparity map image generated using graph-cut 
algorithm dense stereo match. 
Figure 4: image edge map generated by EDISON algorithm 
By using the algorithm and the processes proposed in the paper, 
we design an experiment to prove the efficiency and the 
increased success rate of the building extraction. Currently 
only the box building and hip-roof building been extracted. 
Taking the same stereopair as the input images, compared with 
the ground truth, the accuracy and efficiency of our result is 
comparable to the result of the commercial digital 
photogrammetric workstation , and what’s more, the degree of 
automation is higher than the above two software. 
Due to the self-occlusion and at step edges.( Brenner, Haala 
1998) Thus, the lateral dimension of the building can not 
derive high accuracy, especially in urban areas. The future 
work will be focus on the range image because its high 
accuracy. 
Based on the generation of the DSM data from the stereopair, 
the precise building models are reconstructed by using the 
BMD algorithm. The experiment showed that our method of 
building extraction posses three advantages: the high degree of 
automation, the high success rate and the comparable accuracy. 
Our method can be competed with the optimal algorithm and 
software by virtue of the above advantages. 
REFERENCES 
Baillard, C., Schmid, C., Zisserman, A. and Fitzgibbon, A., 
1999. Automatic line matching and 3D reconstruction of 
buildings from multiple views. In: IAPRS, Vol. 32 Part 3-2W5. 
Baltsavias, E., Gr"un, A., and VanGool, L. (eds), 2001. 
Automatic Extraction of Man-Made Objects from Aerial and 
Satellite Images (III). Balkema Publishers, Netherlands. 
Bauer, J., Kamer, K., Schindler, K., Klaus, A. and Zach,C., 
2003. Segmentation of Building Models from Dense 3D Point- 
Clouds. In: 27th Workshop of the Austrian Association for 
Pattern Recognition. 
Brenner, C., 2000. Towards Fully Automatatic Generation of 
City Models. In: International Archives of Photogrammetry 
and Remote Sensing, Vol. (33) B3/1, pp. 85-92. 
408