The rightmost column represents the results of three models for flat roof reconstruction, and the topmost blue quadrangle in this column shows the roof patch picked up by the user. Note here, the selected roof boundaries are not complete 2-D or 3-D lines. Two of three lines are 2-D lines without any height information and the other one is the 3-D line. After verifying the first 3-D line and selecting the relevant 2-D or 3-D lines, the reconstruction results are illustrated in the second, third and fourth row respectively. It can be very clear to view the results from the blue quadrangles that the reconstruction from the third model is better than other two. The reason is that the third modeling exploits much more information from lines than the junction information used in the first and second model. Certainly non-accurate 2-D lines will make the reconstructed results unreliable. The second and third row at the leftmost two columns (i.e., Test I and II) represent the results of the two slope roof patches. Although one line in the Test I is occluded by the frame of flowers and one line in the Test II is not the real boundary of slope roof patch, the results seem good when verifying by the eyes. The occlusion could be a great problem of roof reconstruction in urban areas, it seems the system entitles the preliminary ability to handle this problem.


Fig.3 The results of roof reconstruction
Fig.4 shows how the system uses the height information of the first 3-D line to correct the wrong 3-D line. The Rightmost means the roof in right image selected by the operator. And the leftmost blue quadrangle shows the result without checking the height information between the first 3-D line and the selected 3-D line during roof reconstruction by the means of the third model. After checking the height information, the right quadrangle (in middle part of Fig.4) is reconstructed correctly.


Fig.4: Illustration of the correction result of wrong 3-D line based on the first 3-D line. Rightmost: the selected roof patch; Leftmost: the wrong result; Middle: the correct result.
Conclusion
Although some drawbacks should be improved in our system, the initial results are satisfactory. Therefore, it proves the framework of this system is feasible. Especially this system really integrates the recognition ability of operator into this semi-automatic system, i.e. the verification of the first 3-D line, the selection of other rooflines and the decision of the final results. This consideration really compensates the recognition ability for the computer in order to handle the complicated imaging environment at the urban areas. The future development for our system is to test for the bigger urban area so as to find the possible problems.

References

• Chio, S.-H., Wang, S.-C., and Wrobel, B., 1999. A Semi-Automatic System for the Reconstruction of Building Roofs in Dense Urban Areas Using Aerial Stereo Images Pairs. AVN ALLGEMEINE VERMESSUNGS-NACHRICHTEN, pp.167-174.
• Cho,Woosug and Shenk,Toni, 1992. Resampling Digital Imagery to Epipolar Geometry. Research Activities in Digital Photogrammetry at The Ohio State University: A Collection of Papers Presented At the XVII Congress of ISPRS, Toni Schenk Editor, Department of Geodetic Science and Surveying, The Ohio State University, Columbus, Ohio, pp.37-43.
• Collgins, R.T., Y-Q. Cheng, C. Jaynes, F. Stollw, X. Wang, A.R. Hanson, and E.M. Riseman, 1995. Site Model Acquisition and Extension from Aerial Images. International Conference on Computer Vision, Cambridge, MA, pp.888-893.
• Fischer, A., T.H. Kolbe, and F. Lang, 1997. Integration of 2D and 3D Reasoning for Building Reconstruction Using a Generic Hierarchical Model. W. Förstner and L. Plümer, Editors, Semantic modeling for the acquisition of topographic information from images and maps. Proceedings of Workshop "SMATI '97", pp.159-180, Bonn Bad Godesberg, Birkhäuser Verrlag.
• Förstner, W., 1994. A Framework for Low Level Feature Extraction. J.O. Eklundh, editor, Computer Vision, ECCV ’94, Vol. II, pages 383-394. Lecture Notes in Computer Science, 801, Springer-Verlag, Berlin, 1994.
• Gruen. A., 1998. TOBAGO-a semi-automated approach for the generation of 3-D building models. ISPRS Journal of Photogrammetry & Remote Sensing 53: pp.108-118.
• Gülch. E., 1997. Application of Semi-Automatic Building Acquisition. A. Gruen, E.P. Baltsavias, O. Henricsson, Editors, Automatic Extraction of Man-Made Objects from Aerial and Space Images(II), Birkhäuser Verrlag, Basel.
• Henricsson, O. and E. Baltsavias, 1997. 3-D Building Reconstruction with ARUBA: A Qualitative and Quantitative Evaluation. A. Gruen, E.P. Baltsavias, O. Henricsson, Editors, Automatic Extraction of Man-Made Objects from Aerial and Space Images(II), Birkhäuser Verrlag, Basel,pp.65-76.
• Heuel, S. and R. Nevatia, 1995. Including Interaction in an Automated Modeling System. Proceedings of the IEEE Symposium on Computer Vision, Coral Gables, Florida, pp.383-388.
• Hsieh, Y., 1995. Design and Evaluation of a Semi-Automated Site Modeling System. Technical Report CMU-CS-95-195, School of Computer Science, Carnegie Mellon University, Pittsburgh, Pennsylvania. Lang, F. and W. Förstner, 1996. 3D-City Modeling with a Digital One-Eye Stereo System. International Archives of Photogrammetry and Remote Sensing, Vol. XXXI, Part B3, pp.415-420, Vienna.
• Lin, C., A. Huertas and R. Nevatia, 1995. Detection of Buildings from Monocular Images. A. Gruen, O. Kuebler, P. Agouris, Editors, Automatic Extraction of Man-Made Objects from Aerial and Space Images, Birkhäuser Verrlag, Basel, pp.125-134.
• McGlone, J.C. and J.A. Shuffelt, 1994. Projective and Object Space Geometry for Monocular Building Extraction. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp. 54-61. Nevatia, R., C.Lin and A.Huertas, 1997. A System for Building Detection from Aerial Image. A. Gruen, E.P. Baltsavias, O. Henricsson, Editors, Automatic Extraction of Man-Made Objects from Aerial and Space Images (II), pp.77-86, Birkhäuser Verrlag, Basel.
• Shufelt, J.A., 1996. Projective Geometry and Photometry for Object Detection and Delineation. Ph.D Thesis (Technical Report CMU-CS-96-164), School of Computer Science, Carnegie Mellon University, Pittsburgh.