International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
3.2 Experiment 2 
In order to further verify the validity of the algorithm proposed 
in this paper, this paper makes another experiment with the 
UltraCamX (UCX) digital aerial images. Firstly, its carries on 
the line extraction using the same method in the Experiment 1, 
and the results is shown as Fig.6. 
    
x 
(left) Target image (right) Searching image 
Figure 6. The line extraction results by improved Hough 
Transform 
Then, the homologous points are obtained by plane-sweeping 
matching algorithm in this paper (Collins, 1995), and the 
homologous points result are shown as Fig. 7. Different with 
the Experiment 1, because of the great elevation differences in 
the image coverage area, for each line, the homograph matrix is 
computed respectively by the homologous points with in the 
neighborhood of each line in the matching process. The final 
matched lines are shown as Fig.8. From the matching results it 
can be drawn that the accuracy rate of line matching is higher, 
but the matching results are relatively sparse. This is because 
there are only few homologous points in the neighborhood of 
lines to be matched, and the computing accuracy of the 
homograph matrix is low. These factors cause to a large 
distortion while projecting to the right image, and hardly find 
out the homologous lines. 
  
et 
Figure 7. Corresponding points are obtained by plane-sweeping 
matching method 
  
> 
(b) Searching image 
Figure 8. The line matching results under the multi-constraint 
conditions 
4 CONCLUSIONS 
The line matching is the hot issue and difficult problem in the 
3D reconstruction. This paper analyzes the technical 
difficulties of this research, and presents the line matching 
algorithm under the improved homograph matrix constraint 
condition focusing on the limitations of existing methods. 
Especially for buildings covered areas in the cities, this paper 
adopts the improved homograph matrix to constraint the line 
matching. For each line to be matched, it computes the 
homograph matrix respectively by the homologous points in the 
line supporting region, and effectively avoids the large 
distortion caused by using the single homograph matrix for the 
image having great elevation differences in the image coverage 
area. This paper simultaneously integrates the multiple 
similarity functions to constrain the line matching, and 
improves the efficiency and accuracy of line matching. The 
deficiency of this paper is the line matching result depending on 
the uniform distribution and intensity of the homologous points, 
and the matching results are relatively sparse. It needs to carry 
on the intensive matching by adopting comprehensive 
homograph matrix or other constraint conditions. 
ACKNOWLEDGEMENTS 
Our research project is supported by the “National Scientific 
Fund Program (No. 40901222, No. 41101452)", the "Open 
Research Fund Program of the State Key Laboratory of 
Information Engineering in Surveying, Mapping and Remote 
  
Intern 
  
Sensing of 
Fund for th 
(No. 20112 
Mosaddegh 
Images. 
bourgogne. 
Fu D., W: 
algorithm 
University 
LT, Liu 
matching s 
173. 
Wu F.C., ! 
homograph 
sinica, 28(€ 
Lou AY. 
dimensiona 
Technical l 
Wu B., X 
textural in 
ISPRS Jou 
pp. 40 - 55 
Wang J.Q. 
based on 
electronics 
Duda, R. O 
to detect li 
pp. 11-15. 
Collins R. 
image mat 
Pattern Re