The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
1370 
5. EXPERIMENT AND RESULTS 
The experiment data is the digital aerial stereo image gotten 
from DMC with scale 1:4000 and 5cm ground resolving power, 
60% overlap, corresponding area is urban and suburban 
combination of plain with abundant terrestrial object. There 
would be 5000-6000 points detected for a single photo when 
the window size is 7X7. About 1025 corresponding points will 
be obtained after character point matching or 794 points after 
epipolar condition detection. The result of relative orientation 
shows that the accuracy of measurement is ±21.6pm that is 18 
pixels size. After least squares matching 774 accurately 
orientation points can be obtained and the accuracy will be 
improved to ±2.3pm, the coarse error is partially eliminated, 
point orientation accuracy has been improved obviously and 
can reach 1/5 pixel. Dense image matching fusing spectrum 
classification result realize the DSM auto extraction on stereo 
model. The buildings and terraces in corresponding area can be 
all denoted as shown in fig 13. Terrain info combined with 
thematic character info obtained through spectrum classification 
can realize the expression and extraction of terrain objects 
effectively. 
integrates point, linear and area character and spectrum classify 
technique to improve the accuracy and reliability of the analyze 
result. The experiment results show that the method introduced 
in paper is effective, but there need further research in such 
fields: improving the holistic geometry restriction in linear 
character and the integrality of multi-images matching. 
REFERENCES 
[1] C. Harris and M. J. Stephens, 1988. A Combined Comer and 
Edge Detector, In Alvey88, pp 147-152. 
[2] Shen Bangle, 1995. Computer Image Processing, PL A 
publishing company 
[3] Wang Zhizhuo, 1979. Theory of Photogrammetry , 
Mapping and Surveying Publishing Company. 
[4] Hu Haiyan, etc. 2007(1). Design of Randomized 
Incremental Algorithm for Terrain Modelling, Geomatic 
Science and Engineering, 
[5] F. Canny, 1986. A Computational Approach to Edge 
Detection, IEEE Trans. Pattern Analysis and Machine 
Intelligence, 8( 6), pp. 679-698. 
[6] Fang Yong, Chang Benyi, 2000(1). The Study of SAR 
Image Map Production Based on the fuse of Multi-Sensors 
Remote Data, Geomatic Science and Engineering. 
[7] Zhang, L., 2005. Automatic Digital Surface Model (DSM) 
Generation from Linear Array Images, Ph. D. Dissertation, 
Report No. 88, Institute of Geodesy and Photogrammetry, ETH 
Zurich, Switzerland, pp.206 
[8] Lukas Z., Andreas K., Barbara G., Konrad K., 2006. 
Towards 3D map generation from digital aerial images. ISPRS 
Journal of Photogrammetry & Remote Sensing , 60(6), 
pp.413-427. 
Figure 13, The 3D Play for Auto Extracted DSM 
6. CONCLUSIONS 
An effective method is put forward in the paper for resolving 
the difficulty be confronted with high resolution stereo image 
auto processing. The main target is to resolve the problem that 
the stereo matching success ratio falls as the resolution 
improving. The main cause is that with the resolution improved 
the ratio that character indistinct rise, which leads to big 
difference between matching result and factual situation. The 
paper introduces a strategy fusing multi characters which