International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
For areas of high relief such as Great Smoky Mountains
National Park, Blue Ridge Parkway and Cumberland Gap, the
overlays must be differentially rectified using a DEM to
remove the effects of relief displacement, which at times can be
quite significant (see Jordan, 2002). Improper corrections can
lead to major difficulties in edge matching detail in the overlap
areas of adjacent photographs along a flight line. The
mountainous terrain in Great Smoky Mountains National Park
is the source of major relief displacements in the large
(1:12,000) scale aerial photographs. These relief effects greatly
influence the apparent shapes of objects appearing on adjacent
photos as well as their map positions and areas. Thus, it is
important that the polygons are corrected properly in shape and
position to facilitate edge matching during its incorporation
into the GIS database. For example, a distinct area appearing
on the aerial photographs in the Thunderhead Mountain area in
the central portion of the park near the Appalachian Trail
occurs on a steeply sloping mountainside. Elevation ranges
from 1549 m in the lower left corner of the image chip to 1214
m in the upper right — a range of 335 m over a distance of about
600 m. When viewed on the three overlapping photographs,
the area appears to be vastly different sizes and shapes (Figure
3). Thus, mapping the area from each of the three uncorrected
photos would potentially give different results.
s
P) Nd Ab) (c)
Figure 3. The dark shadowed area in the above image chips
appears to be very different in shape and size in these three
overlapping photographs. The image chip (a) is from the lower
right corner of Photo 10063; b) near the bottom center of Photo
10062; and c) lower left edge of Photo 10061.
COMPARISON OF RECTIFICATION METHODS
There are a number of well-known image rectification methods
available that can be used for converting vegetation overlays in
raster format to a vector map base. Three of these are 1)
polynomial (affine) based on a least-squares fit to two-
dimensional GCPs; 2) single-photo projective rectification
referenced to a mean datum elevation using a photogrammetric
solution and 3-D GCP coordinates; and 3) rigorous differential
correction (orthocorrection) using the photogrammetric
solution and a DEM (Novak, 1992; Welch and Jordan, 1996).
To compare the effectiveness of the techniques, Photo 10063
from Thunderhead Mountain was rectified using each of the
three methods and then overlaid with the completed vegetation
map (Figures 4a-d). In the following examples, the darker
shadowed area and corresponding vegetation polygon indicated
by the black arrow in Figure 4a will be used to illustrate the
effects of the different rectification methods. In the GIS database,
this polygon has an area of 5.97 ha (Table 2).
After aerotriangulation, 14 GCPs were available for Photo 10063.
The affine transformation coefficients were computed using the
method of least squares and resulted in an RMSE at the 14 GCPs of
106 pixels or 53 m. Most of this error is due to relief displacements
in the image. The aerial photograph was then rectified using the
polynomial method. The resulting image is approximately in the
correct geographical location but relief displacements have not been
corrected (Figure 4a). Although the general correspondence
between the vegetation polygons and the underlying image can be
seen (point A on the photo), it is clear that the overall registration
accuracy is poor: the lines from the vegetation coverage do not fit
this rectified air photo well and the shape distortions in the image
are clearly visible. In this case, the dark shadowed area in the photo
corresponding to the polygon (indicated by the arrow) appears to be
longer, wider and in a different position than the actual polygon in
the vegetation coverage. In this figure, the polygon measured
directly from the image has an area of 8.34 ha, which is 2.4 ha (40
per cent) greater than the actual area of the polygon taken from the
GIS database.
The overall geometry of the image rectified using the single photo
projective transformation was not improved significantly over the
polynomial rectification (Figure 4b). The photogrammetric solution
used to determine the exterior orientation parameters, however, was
excellent and yielded a RMSE of 3.34 pixels or 1.67 m at the 14
GCPs. The image was then rectified to an elevation datum value of
1380 m using a method which enforces the scale at the datum and
corrects for tilt but does not correct for relief effects. Note that
although the vegetation polygons generally do not fit the image
exactly, there is a good fit in the areas near the 1380 m contour
(shown in yellow) where scaling is exact using the photogrammetric
solution. Overall, the shapes of the target polygon and other
features are still distorted and this solution is not satisfactory. The
area of the sample polygon measured from this image is 7.9 ha.
Orthocorrection was performed on the photo using the same exterior
orientation parameters computed above, but this time using the
USGS DEM to provide elevation values to correct for relief
displacement at each pixel location (Figure 4c). Polygons in the
completed vegetation coverage are aligned perfectly with the
underlying orthophoto (see point A) and the shadowed area
indicated by the arrow has an area of 5.98 ha which corresponds
well with the value in the GIS database for the polygon. This high
level of correspondence clearly demonstrates the requirement for a
full softcopy photogrammetric solution to rectifying vegetation
overlays.
Finally, as a logic check, the vegetation vectors were overlaid on the
USGS DOQQ (Figure 4d). It is reassuring to see that the GIS
database created by orthocorrection techniques described in this
paper lines up very well with the USGS DOQQ product of the same
area.
660
Inte
Tal
