— 10m 
  
  
  
] p^. d] 
i 
  
  
Fig. 2: Error vectors in X and Y using a GPS accuracy 
of 1.7m on the ground and 1.4m at the perspective 
centers. (the errors are in meters, ground units, at all 
points) 
  
  
— 1.0 ma 
  
  
  
Fig. 3: Error vectors in X and Y using a GPS accuracy 
of 1.7m on the ground and 1.4m at the perspective 
centers (the errors are in meters, ground units, at all 
points) with unknown interior orientation parameters. 
  
Fig. 4: Error vectors in X and Y using a GPS accuracy 
of 1.7m on the ground and 2.0m at the perspective 
centers. (the errors are in meters, ground units, at all 
points) 
In summary, the linear feature only overcomes 
the datum deficiency due to the inability of solving for 
the roll angle. Thus, it will contribute nothing towards 
computing the other orientation parameters. 
5. Results with Real Data 
This technique was applied to triangulate strips 
of digital aerial photos captured along a railroad line. 
This project was conducted by the Center for Mapping 
of The Ohio State University together with Ruekert and 
Mielke (Waukesha, WI). It was sponsored by Conrail 
(Philadelphia, PA) and MCI (Houston, TX). GPS 
observations were captured along the rail road by 
loading the Center's GPSVan on a flat car, which was 
pulled by a locomotive for 50 miles (figure 5). The van 
collected both GPS positions to define the rail 
centerlines, as well as digital image pairs and analog 
videos. Digital aerial images were collected by The 
Sewall Company (Old Town, Maine) using a Kodak 
DCS camera (1280x1024 pixel frame CCD sensor) and 
a Trimble 4000ST GPS receiver. The results of GPS 
controlled strip triangulation (tie point coordinates) 
were used to rectify the aerial images to generate digital 
orthophotos of the railroad tracks and the surrounding 
areas. 
In this project the GPS positions at the aircraft 
were computed with a standard deviation of 3.5 m. The 
rail-centerlines are accurate to about 1.5 m. Image 
coordinates of 10 to 15 points per model and 5 to 6 
points along the linear feature were measured 
monoscopically on a computer monitor. We estimate 
that the measurement accuracy is better than 1 pixel 
which corresponds to 30 cm on the ground. Using these 
image coordinates the polynomials were interpolated 
and the bundle adjustment was computed. The variance 
component was typically two to three pixels large. All 
together 21 strips of about 8 photos each were 
triangulated. 
  
Fig. 5: GPSVan loaded on a flat-car, Stereo-vision 
system. 
208 
  
The resulti 
6: the rail 
super-impc 
images. 
6. Co 
A 
triangulati: 
presented. 
data. The 
proposed n 
of ground 
mobile m 
Mapping's 
also be eqi 
feature col