International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
  
elevations have minor effects. The true terrain elevation can be 
obtained from stereo images with sub-meter accuracy using 
refined RPCs. 
3.1.3. An Example Of A 3D City Model. In order to 
demonstrate the ability of the above-described RFM scheme to 
quickly and efficiently generate 3D city models, a portion of 
Downtown San Diego was mapped with a single IKONOS 
image and a DEM from the data set that was described. Each of 
the buildings was modeled interactively by measuring its height 
using the 3D floating cursor technique and digitizing the 
contour of its roof. Figure 4(a) depicts an example of a 
combined 2D and 3D data collection. Figure 4(b) depicts a 
photo realistic view of a draped ortho-image over the DEM and 
the 3D city model. 
  
  
  
  
(a) 
  
  
Figure 4. Mapping results with SilverEye: (a) 2D vector data 
collection; (b) a 3D visualization of a single image 3D mapping 
of downtown San Diego. 
3.2 Results Using the RFM For Aerial Imagery Processing 
The RFM scheme can be applied to aerial imagery using both 
the terrain-dependent and the terrain independent solution, To 
demonstrate this, an aerial stereo pair of the Ottawa city area 
was processed. The stereo pair covered an overlapped area of 
about 17 km?. Both images had a ground resolution of 0.24 m, 
and were provided with the entire interior and exterior 
orientation parameters, the camera calibration data including 
fiducial marks and parameters describing the radial lens 
distortion. The flying height of the airplane was 2400 m above 
MSL. For each image, a second-order RFM was computed by 
the terrain-independent approach in SilverEye. The worst errors 
of the RFMs were 0.002 pixel in line and 0.0018 pixel in 
sample at the check grid points. 
A topographic map sheet with the scale of 1:1250 of the study 
area of Ottawa city with 2-feet contours was also used. This 
map included the elevations of the building roofs, which are 
also annotated on the map (in feet). The map was compiled 
from aerial photography flown in May 1971 using the 
traditional photogrammetric processing techniques. The vertical 
datum and horizontal datum are Geodetic Survey of Canada 
(GSC) and North American Datum of 1927, respectively. 
Based on this data, the heights of 12 buildings were obtained in 
three different ways (Table 4): 
+ First, the building heights are read from the map. The 
elevation of the base is interpolated from the map 
contours. However, the building base points are usually 
found to be higher than the surrounding terrain. Therefore 
these effects should be removed by subtracting the relative 
height of the base point relative to the terrain surface. 
. Second, the building height is also measured using the 
projection utility on the right image. This was done by 
raising the cursor from the base point to the building roof 
along the side of the buildings. Because we do not have a 
DTM, the average terrain elevation (82.6 m MSL) is used 
to approximate the elevations of the building footprints. If 
the average elevation is increase to 120 m MSL, then the 
measured heights will be 0.51 m smaller in average than 
those listed in Table 4. While the mean error is 0.34 m for 
lkonos as shown in Table 3. So this effect is more 
significant for aerial images than satellite images. 
« Third, the elevations of the building base, the te 
the roof were measured using stereo pair. Th 
elevation is intended to eliminate the effect of higher base 
over terrain when compared with the figures read from the 
map. Consequently, we measured the height correction 
(the last column in Table 4) of the base point relative to the 
natural terrain using the stereo pair to eliminate this 
systematic bias of heights since such data cannot be read 
from the map. These height corrections are all positive 
values, and this show that the building footprints are 
usually higher than their surroundings. 
rrain and 
e terrain 
The differences between the building heights obtained using the 
above three ways are compared in Tables 5 and 6, respectively 
with and without systematic biases. It can be observed that the 
largest difference is 0.9 m only when comparing the stereo- 
based heights and the corrected map heights. Overall, as can be 
observed from Table 6, the heights measured using stereo 
images were more accurate than those using a single image. 
4. CONCLUSIONS 
The  RFM framework provides a comprehensive 
photogrammetric solution in a variety of applications. It offers 
greater flexibility and enables non-technical users to exploit the 
full potential of high-resolution imagery. Using this framework, 
users are able to overcome two traditional barriers in 
photogrammetric processing, namely the requirement for a 
physical sensor model and the requirement of triangulation 
using GCPs for deriving the sensor orientation. 
In contrast to this, it should be noted that currently the 
implementation of the RFM scheme and its adaptation in 
practice heavily depends on data vendors. As users are not 
provided with tools to generate their own RPCs, their ability to 
adopt and utilize the RFM framework depends on the 
availability of RPCs that are supplied with the raw imagery 
data. Yet, as this new framework is being rapidly adopted for 
1026 
Cr 
20( 
dat 
Ma 
Fra 
fun