Aluir Dal Poz
Six grouping of straight lines were selected from building edges using an existing DXF file (1:2000), generated by analytical
photogrammetric plotting. Selection of object straight lines was done considering the worst case, i.e., only a few entities for
each grouping was collected. In real operation environment much more entities could be selected, improving the final
results. The corresponding area of the selected features was automatically searched and the straight lines in the image-space
were automatically extracted. Figure 4 shows an example of the feature extraction results and the final results of the
matching process for one grouping. Table 1 presents the number of entities both in the image-space and object-space for the
selected groupings.
The approximate and refined exterior orientation parameters and their standard deviation are presented in the table 2. Null
values were adopted as approximate rotations, with standard deviations of 0.052 rad (3°). On the other hand, the co-ordinates
of perspective center were measured by GPS and used as approximate translations. Precision of 30m for X and 10m for Y
and Z-axis were assumed. After processing, using the proposed methods, 3 correspondences from grouping 1, 2 from
grouping 2, 1 from grouping 3, 1 from grouping 5, and 1 from grouping 6 were obtained. Therefore, the orientation
parameters were estimated with 8 correspondences.
Approximate Values refined Values
Parameter Standard Parameter Standard
Values Deviation Values Deviation
K rad 0 0.052 -0.06988 0.00042
¢ rad 0 0.052 -0.00261 . 0.00235
o rad 0 0.052 0.02527 . 0.00341
X,m 664402 30 664380,9 3,9
Yım 7481112 10 7481115,9 5,1
Zo M 1334 10 1329.0 2,8
Table 2. Approximate and refined exterior orientation parameters and
theirs standard deviations
A conventional space resection process using ground control points was performed in order to check the results of the
proposed approach. The differences (£) between parameters of a conventional space resection and the proposed approach
are presented in table 3. Errors for parameters K, «y, X,, and Z, were greater than the respective standard deviations that are
estimated, which was probably caused by small and nearly grouped straight lines, generating weak geometry.
Errors in the exterior orientation parameters
£K EQ £ (0 £ X, £Y, eZ.
(rad) (rad) (rad) (m) (m) (m)
-0.0111 -0,0051 0,0009 -8,5 2,7 -4,9
Table 3. Errors obtained in the estimation process
Finally, using translation and rotation errors that is obtained in each correspondence, the convergence of IEKF in the
matching process is presented in the figure 5.
The following conclusions can be derived from this figure:
* The selected groupings were not good enough to provide a high precision solution;
* IEKF starts to converge after the second correspondence, except in K. This can be explained by the weak geometry
provided by a few and small straight lines; and
* Even with this weak geometry all the correspondences were correctly established and the filter converged properly.
212 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.