Chunsun Zhang
process is repeated and produces a tree of possible subdivisions. Then, unwinding the recursion back up the tree, a
decision is made at each junction point as to whether to replace the current low level description with a single higher
level segment by checking the connecting angle.
3.4 Experimental Results
The proposed method for straight line matching is implemented and experiments have been performed on a number of
areas extracted from aerial images. The test areas cover different terrain and landcover, including rural areas, suburban,
urban, and hills. Because of a limited space, the authors describe one of the dataset used, and report and analyze the
major results. The dataset (Fig. 6) is extracted from a stereo pair at a test site of the ATOMI project. The line extraction
process resulted in 985 and 971 straight lines in the left and right images respectively (see Fig. 7). Only 789 and 809
lines in the left and right images are longer than 5 pixels. 460 matches are found, of which only 5 are wrong matches.
The matching result is shown in Fig. 8.
Figure 6. The test image pair. Figure 7. Extracted straight lines. Figure 8. The matching results.
4 DISCUSSION AND CONCLUSIONS
We presented a new scheme for road reconstruction from aerial images. The proposed idea uses as much information as
possible to increase success rate and reliability of the results. As one of the key components of the system, we presented
a method for 3D line generation through stereo matching. The matching approach has high success rate and most
importantly is very reliable. It makes use of rich attributes for matching, including line geometrical properties and line
flanking regions photometrical and chromatic properties. This is an advantage over other approaches that only use line
geometry and line gray scale information. The developed structural matching method achieves locally consistent results,
allows matching in case of partially occluded edges, broken edges etc. The use of similarity scores priori to structural
matching greatly speeds up the process. Although used here for straight edges, this method can be easily extended to
arbitrary edges, or even points, if some of the matching criteria (feature attributes) are excluded or adapted.
Besides the work on straight line matching and 3D line generation, we completed a multispectral image classification
method to find road regions. Guided by the initial knowledge base, we excluded the lines outside the road buffer area
(this area is defined using the road centerlines of VEC25 and their estimated maximum error). By combining the 2D
lines with the classification result, a relation with the road region (in, outside, at the border) is attached to each line.
Lines with a slope difference to the slope from the known local DHMO5 larger than a certain value are excluded. Our
future work will focus on the extraction of roadsides by fusion of various 2D and 3D information extracted from images
and knowledge from a road database, and modeling roads of different classes in various terrain relief and landcover. As
reliability is the most important figure in our system, the system will use as much as possible information to create
redundancy. Cues like cars on road and road marks will be extracted to confirm the reconstruction results. Finally, a
metric for quality estimation will be developed, and the results will be evaluated with manually extracted ground truth
reference data.
ACKNOWLEDGEMENTS
We acknowledge the financial support of this work and of the project ATOMI by the Swiss Federal Office of
Topography. The discussions with IGP members and members of the projects ATOMI and AMOBE are gratefully
appreciated. The first author would also like to thank Prof. A. Gruen for his guidance and comments.
1014 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.