Devin Kelley 
  
The following processing settings were applied for all of the simulated data sets: 
e The threshold O for terminating the iteration process was set to 1.0E-7. 
e The approximations of the tie points were displaced from their actual positions by approximate values of 100m 
(horizontally) and 10 m (vertically) to verify their adjustment. 
  
  
  
FRAME IMAGERY Without linear features With linear features 
Rmsx [m] 0.024 0.040 
Rms y [m] 0.031 0.034 
Rmsz [m] 0.106 0.071 
  
  
Table 2: Rms-values of the bundle adjustment of frame imagery 
  
  
  
THREE-LINE SCANNERS Without linear features With linear features 
Rmsx [m] 2.639 1.688 
Rms y [m] 1.292 0.825 
Rmsz [m] 0.550 0.559 
  
  
Table 3: Rms-values of the bundle adjustment of three-line scanners 
  
  
  
PANORAMIC LINEAR ARRAY Without linear features With linear features 
SCANNERS 
Rms x [m] 0.522 0.376 
Rms y [m] 0.797 0.300 
Rms z [m] 0.935 0.553 
  
  
Table 4: Rms-values of bundle adjustment of panoramic linear array scanners 
As illustrated in Table 1, the straight-line constraint did not improve the RMS values in the case of frame imagery. 
However, in the case of linear array scanner imagery, there was a noticeable improvement. With linear array scanner 
imagery, many EOPs are involved in the adjustment, and the added equations constrain the solution, aiding in the 
determination of the EOPs. Therefore, it is advantageous to evaluate as many image points along the straight line as 
possible. 
5. CONCLUSIONS/RECOMMENDATIONS 
A new approach was developed to handle object space straight lines in linear array scanner imagery. Because of the 
nature of line cameras, straight lines in object space may not appear as straight lines in the captured scene. In the 
proposed constraint, object space lines are defined by two points, which must be identified in at least one scene. 
Using this technique with linear array scanner imagery, one independent constraint equation is added to the adjustment 
for each image point evaluated. The added constraint equations aid in the recovery of the many exterior orientation 
parameters associated with linear array scanner imagery. It is therefore advantageous to evaluate as many image points 
along the straight line as possible. This constraint is also valid in the case of frame imagery. Also, the incorporation of 
this constraint into existing bundle adjustment software is straightforward. 
Testing with simulated data proved the superiority of this technique over aerial triangulation with disconnected points. 
Recommendations for future work: 
e More testing with real data. We would like to use an available GIS database and data collected by terrestrial mobile 
mapping systems, e.g. road networks, to provide control for aerial triangulation. 
e Automatic extraction and matching of linear features from imagery. 
  
184 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 
Ha 
Sc 
Un 
Ku 
Re 
AS