245 
by correlation. The flight plan can be adjusted to 
minimize their impact in two ways: modifying the 
overlap and carefully considering terrain morphology. 
Quoting Ackermann, in digital photogrammetry 
“redundancy should replace intelligence”: taking and 
processing several images of the same area rather than 
a single model is therefore acceptable, if it pays off in 
terms of better overall accuracy, reducing the number 
of mismatches. The new flight has therefore been 
designed with 80% overlap and 60% sidelap, in order 
to derive models with a less favourable height to base 
ratio, but with smaller perspective distortions. They 
will be processed first, to provide an initial DEM to be 
later refined with the 60% overlap models. 
Using sidelap up to 60% yields four-fold model 
coverage of the same terrain patch. The advantage here 
is that each model has a different viewing direction, 
increasing the possibilities that occluded areas are 
visible in at least one model. Currently, virtually all 
DEM generation software is stereo-based; thought the 
use of several images is envisaged and now generally 
recognized as beneficial, this can be implemented in 
two significantly different ways: repeating DEM 
generation in different stereomodels over the same area 
is not as effective as actually performing DEM 
generation with all images at the same time, like in 
MGCM (Gruen and Baltsavias, 1985) or similar 
approaches. Using several models yields elevation data 
at nominally coincident grid knots. The question 
therefore arises what is the best weighting of each 
elevation value at a given location. A simple averaging, 
thought optimal in most cases, would certainly fail in 
problem areas, where more than one value may be 
grossly falsified. Even more robust estimation 
methods, with a possible outliers percentage larger than 
50%, may give incorrect results. 
As far as flight direction is considered, if the 
excavation front of the quarry is elongated and follows 
roughly the contour lines, then mimimization of the 
occlusions may be achieved by using only half image 
and flying just over the lower edge of the front. Parallel 
strip may be flown, if necessary, maintaining a 60% 
side lap, covering the quarry from the lower edge 
upwards. 
Illumination may affect the results, given the strong 
terrain reflectance, either because of the shadows as 
well as for the dynamic range of the scene. 
Minimization of the disturbances is achieved by flying 
at midday and/or prescribing a flight in a cloudy day. 
Good image quality make it necessary a FMC system: 
its effect is nevertheless limited because of the large 
heigth differences in the model, which leaves a 
significant residual motion. 
2.2 Reference data 
The contour lines file of the 1997 map 1:1000 has been 
made available by the Comune di Botticino. The 
accuracy of elevation data therein may be assumed to 
range from 20 to 50 cm. Data have been interpolated 
either by Kriging and by Delaunay triangulation over a 
2m grid, to be used to compute the differences with the 
elevations estimated by the DPWs. In order to provide 
a common reference system for the comparison, we 
used the ground coordinates of the tie points from the 
analytical AT of the strip, that is just the same points 
used for the absolute orientation in the map restitution. 
A topographic survey with a motorized laser scanning 
theodolite will also provide (sparse but precise) 
reference data for the second flight. 
2.3 Using orthoimages to check a DEM 
To provide an additional check of the accuracy of a 
stereo DEM, a technique may be used which employes 
two orthophotos, derived from each image of the pair, 
by using the DEM to be controlled. By comparing the 
(nominally fully) corresponding images so obtained, 
errors in the DEM can be traced according to 
(Baltsavias, 1996). The underlying concept is depicted 
in Fig. 1: if you select a feature in one orthophoto, its 
counterpart in the other should show up at the same 
(X,Y) location in object space. If not, this is because 
the DEM elevation in the area is wrong. The correct Z 
value may be determined (and so the error computed) 
by going back to the image space and intersecting the 
correct pair of homologous point. In digital 
photogrammetry this technique is very simple to 
implement and automate: extensive testing of DEM 
quality (in principle the whole DEM may be checked) 
is therefore 
Fig. 1 - DTM check by using two orthophotos 
feasible. Once the two orthophotos have been 
generated, feature points may be selected with an 
interest operator or on a grid. Image correlation 
techniques can be employed to find the homologous in 
the second orthophoto, taking advantage of the 
nominally coincident initial location and of the smaller 
perspective distortion that the orthophoto should 
display, compare to the aerial image. If the discrepancy 
between the matched positions, converted into object 
space, are within tolerance, the DEM is correct. 
Otherwise, each orthophoto point is back projected into 
the original image and the two rays are intersected, 
yielding the correct ground coordinates. 
2.4 Image scanning 
Up today image scanning has been performed only on 
the 50_51 model of the 1997 flight. Images have been