Kerry McIntosh 
4.0 Data Fusion 
Edge pixels were extracted from a stereopair of aerial images using the optimal zero-crossing operator (Sarkar and 
Boyer, 1991). Figure 3 shows the edge pixels as an overlay on the aerial image. The area depicted is the region used 
for the experimentation discussed in this paper. 
The segmentation of the edge pixels creates straight line segments. Only segments longer than ten pixels were used in 
the segment matching process. The segments in the left and right images were searched to find corresponding 
segments. The segments shown in Figure 4 are those that were found to have a matching segment in the second image. 
The line segments were reconstructed to three dimensions using the information available from the interior and exterior 
orientations of the aerial imagery. A plan view of the reconstructed line segments is shown in Figure 5. As a 
comparison, breaklines associated with the buildings were measured manually and are shown in Figure 6. Future 
research will investigate using multiple images to provide higher accuracy line matching, as presented by Schmid and 
Zisserman (1997) and Baillard et al., (1999). The inclusion of this approach will depend on the availability of multiple 
images over the area of interest. 
The laser data have been classified according to elevation, and are shown in Figure 7. The manually measured 
breaklines were included in this figure to illustrate the effectiveness of the classification procedure for this data set. The 
surface created using only the laser points is shown in Figure 8. This serves as a comparison for the surfaces created 
using the combined data sources. 
The surface generated using the laser data and the manually measured breaklines is shown in Figure 9. This is the 
representation of the visible surface that the research attempts to achieve automatically. Figure 10 shows the surface 
generated using the laser data and the automatically extracted breaklines, with breaklines added at ground height. A 
visual comparison of Figures 8 and 10 shows the improvement achieved by including the automatically derived 
breaklines with the laser data when determining the visible surface model over the urban area. 
mo 
rentrer 
   
  
  
  
    
  
  
  
  
  
suo —— : of + 7 : 
2025 + j ET nC | | 2025 } | 
B Tg °° zd 2629] | 
2015} | [N : / / m ee H 2015. | {| 
2010} } | — ue aw 200 | (0——— 2) E 
2005 + I | ki = = IT == E 2005} | | f eee f Lf IL /| — 3 ] 
pH j | j P LL Wl ili ] 
Ba; tet el Gute 7 haa! elfe ] 
19051 tu | Sli ; hid 1995 L. aet M ri | 
1900} - ; Lr ud 10001 od EL 
D oem NY mm pe app 1086 : | 
1980 ; : on db 0 Toe ia p] 
4830 4890 4900 4910 4920 4930 4880 4890 4900 4910 4920 4930 
Figure 5. Reconstructed line segments. Figure 6. Manually measured breaklines. 
  
568 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
203 
202* 
202 
2016 
201 
2005 
2000 
This 
sche 
inte; 
visit 
Furt 
info: 
data 
ACI 
This 
REF 
Ack 
Phot 
Axel 
Digi 
Axel 
and | 
Baill 
builc 
Balt: 
and ]