——— ts P, 
pe 00 CRMDERNEENRUELL DIL anc 
ied 
ed 
International Archives of the Photogrammetry, Remote Sensing 
3.1 Results With Ikonos Imagery 
The test was carried out using a sample panchromatic Ikonos 
stereo pair (ortho-kit product level) that was provided by Space 
Imaging, and a 30m USGS DEM. First, the RPCs supplied with 
the images were refined with three high-accuracy GCPs using 
the refinement method described in section 2.2. The RMS errors 
at these GCPs before and after the refinement are given in 
Table 1. Then, using the above data, 33 well-distributed 
horizontal distances were measured using a single image and a 
stereo pair. In addition, measurements of the same distances 
were carried out in stereo using two leading commercial 
photogrammetric packages: PCI Geomatica© and ERDAS 
ImagineO. Both packages support the RFM. 
  
  
  
Ikonos left Ikonos right 
Line Sample Line Sample 
Original RPCs 5.37 5.34 5.21 5.04 
Refined RPCs 0.57 0.12 0.57 0.32 
  
Table 1. RMS errors in image space (Unit: pixels) 
3.1.1 2D distance Measurement. An analysis of the 
Horizontal distance differences was carried out as the 
following: first the single image (“mono”) measurements were 
compared to the stereo pair measurements that were obtained in 
PCI Geomatica© and ERDAS Imagine© (Figure 3a and 3b). 
Then the performance of the stereo processing scheme was 
compared to the stereo pair measurements that were obtained 
from the two software packages (Figure 3c and 3d). In addition, 
for each comparison a Gaussian curve was fitted (red curves in 
Figures 3a through 3e). 
The summary statistics of the various comparisons are given in 
Table 2. The residuals that were obtained are bellow 1.5 m in 
all the cases as can be observed from Figure 3. No systematic 
effects were detected in all cases. It should also be noted that in 
both comparisons of mono to stereo pair measurements a higher 
standard deviation was obtained than in both comparisons of 
stereo measurements in SilverEye to stereo measurements in the 
other packages. 
The largest standard deviation is smaller than 0.8 m for these 
comparisons. And this shows a sub-meter relative accuracy 
when measuring object dimensions using the RFM. It is 
interesting to note that the residuals that were obtained between 
SilverEye to PCI and ERDAS (Figures 3c and 3d) in the stereo 
measurement mode are of the same magnitude as the residuals 
that were obtained between PCI and ERDAS in stereo mode 
(Figure 3e). 
  
Silver Eye Silver Eye Silver Eye Silver Eye 
  
(mono) (mono) (stereo) (stereo) 
vs. Vs. Vs. Vs. 
PCI ERDAS PCI ERDAS 
(Stereo) (Stereo) (Stereo) (Stereo) 
Mean 0.09 0.12 0.03 0.03 
Std +0.50 +0.49 +0.70 +0.72 
Min -0.85 -0.84 -1.23 -0.72 
Max 1.14 1.13 0.90 0.62 
  
Table 2. Summary statistics for mono and stereo 2D 
distance measurements (Unit: meters) 
and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
INI 4 I 
(a) (b) 
(c) (d) 
  
M, 
(e) 
Figure 3. Evaluation of the distance difference: (a) SilverEve — 
mono vs. PCI — stereo; (b) SilverEye — mono vs. ERDAS 
Imagine — stereo; (c) SilverEye — stereo vs. PCI — stereo; 
(d) SilverEye — stereo vs. ERDAS Imagine — stereo 
(e) PCI — stereo vs. ERDAS stereo. 
(In all figures: x axis represents residuals in meters and y axis 
represents the residual frequency) 
3.1.2 Height Measurement. A second test was carried out in 
order to evaluate the performance of the single image (*mono") 
and stereo-based height measurement in SilverEye. A set of 22 
buildings were selected in the downtown San Diego, and four 
measurements were carried out for each building: mono and 
stereo height measurements with SilverEye, and stereo height 
measurements in PCI GeomaticaO and ERDAS Imagine©. The 
results that were obtained are summarized in Table 3. 
  
Silver Eye Silver Eye Silver Eye Silver Eye 
  
(mono) (mono) (stereo) (stereo) vs. 
Vs. Vs. Vs. ERDAS 
PCI ERDAS PCI (Stereo) 
(Stereo) (Stereo) (Stereo) 
Mean 0.34 0.34 -0.45 -0.45 
Std +1.54 31.53 +0.75 +1.06 
Min -1.37 -1.74 -2.26 -2.12 
Max 4.27 4.45 1.13 1.92 
  
Table 3. Summary statistics for mono and stereo building 
height measurements (Unit: meters) 
In this case largest residuals (above 4.0 m) were obtained for 
the mono measurements in comparison to the sterco 
measurements (bellow 2.0 m). However as can be found from 
Table 1, the standard deviation for single images is comparable 
with that for stereo images. This is because the orbit height of 
satellites (approximately 680 km) is significantly larger than the 
object heights and the shifts of terrain elevations of the building 
footprints in the coarse DEM relative to the true terrain 
1025