Zheng Wang 
  
twenty-five contours, including one building, were classified as non-building contours for various reasons: three of 
them had length problems, either too long or too short; two of them were not closed; four of them were nonsymmetric; 
fifteen of them had too large circularity values; and one of them was acturally a pit. The misclassified building was 
caused by nearby trees. Several observations can be made from the statistics. First, the results in table 1 demonstrates 
that the shape and generic condition-based analysis and classification was very effective to get rid of many non- 
building objects. In the total forty-one contours, eight of them were true building contours and the remaining thirty- 
three were non-building objects. This process successfully elliminated 75 percent (twenty-five out of thirty-three) of the 
total non-building objects. Second, fifteen out of total twenty-five rejected contours were caused by circularity problem. 
The circularity seemed to be more effective than symmetry to separate building contours from non-building contours. 
Third, after the symmetry and circularity classification half of the classified building contours actually belonged to non- 
building objects. But, all those non-building contours had the same high symmetry and low circularity as the true 
building contours, which means symmetry and circularity alone were not enough to completely divide buildings and 
non-building objects. To complete the classification, further filtering based on orthogonality and parallism was applied 
to the remaining contours. At the end, eight final building contours were extracted, which are shown in figure 6 as 
contours marked with *. 
Bldg Contour Length Closure Symmetry — Circularity Pit 
| 16 [5 | 2 | 4 I$ [1] 
Table 1. A statistics of 41 contours 
  
After building contours were classified, they were used to extract building points. Following the procedure described in 
section 2, TIN models were generated and building surfaces and corners were derived. Figures 7-14 show the 
reconstructed buildings together with their corresponding TIN models. Please note that the buildings are plotted with 
different scales. 
  
Figure 9. Corners, structure and TIN of Building 10. Figure 10. Corners, structure and TIN of Building 11. 
  
962 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.