Cruz, Santiago
Two additional cases are shown in figure 7. These examples show cases where weak conditions for the application of
this method can be found. The first case (A), although it was successfully solved, corresponds to a photograph of a soil
(taken from Carvajal, 1999). The camera is mounted on an aluminum frame that allows near vertical stereoscopic pairs
from 2 meters height. Because there are targeted control points at the base of the aluminum frame, it is possible to
obtain the digital elevation model of the soil surface. The comparison between different DEMs can permit calculation of
soil losses as well as detailed studies of the micro-relief. These kind of photographs can have problems for the
automatic inner orientation as the frame structure can be confused with the edge because that is contrasted enough and it
appears near parallel to the edges. Also contrast is very weak at the lower corners because poor illumination conditions.
At the center, the lower edge is clearly detected because is marked by the edge frame and the aluminum rod, but near
the corners the aluminum frame (largely affected by distortion) is more contrasted than the edge frame. Gradients are
higher at the aluminum frame than at the format edge. Because the measurements in the aluminum frame at the corners
have high residuals they are eliminated and they don't contribute to the line fitting. Line is only fitted with the
measurement at the center of the format where the edge frame is clearly visible. However this situation not always is
properly solved if the aluminum frame had been pictured clearly separated (but very close) to the lower edge. In the
upper edge, because the aluminum frame is more separated from edge, it is possible to avoid this situation when the
search areas are defined.
The second case at the figure 7 (B) shows a frame taken with the Hasselblad camera. The lower right corner has not
been properly computed (it appears displaced upward the true position, B2) and high residuals are obtained (table 1).
Some white lines (from the road in the lower part) appear very close the lower edge format throughout the edge (such as
the white line at the right corner, B1). These lines have a strong influence in the regression fit line and it has been
impossible to adjust automatically the correct position of the lower corners. Finally, the inner orientation of the frame
was made manually and the residuals were lower than the pixel size (42 um).
Al
A: 35 mm (1200 dpi)
B: 70 mm (600 dpi)
Bl B2
Figure 7. Weak conditions for application of the automatic inner orientation.
Example A: Successfully solved, but similar situations can cause problems.
Example B: Unsuccessfully solved, white lines of the limit of the road, just at the lower format edge (B1),
displace the computed corner upward the true position (B2).
4 CONCLUSIONS
A program for automatic inner orientation of digitized images taken with non-metric cameras (35 and 70 mm) has been
presented. This program, under IDL 5.0, detects the format edges and computes the regression fit lines that define the
frame. Frame corners are computed from intersection of lines as well as the indicated principal point (intersection of
diagonals). Routines for coordinate transformations (4 and 6 parameters, two-dimensional projective and bilinear) has
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 155