181 
A check point analysis which photogrammetrically compares computed ground coordinates of tp’s with 
coordinates measured indipendently by geodetic methods was also performed. The result (in km) is splitted into 
the three components - see Table 3. 
The matching process between the two stereo images is area-based; it requires some corresponding points 
in the overlapping area of the two images and a coarse grid defined in the left image. A coarse match is performed 
using the image-pyramid model (four times the same image at different spatial resolution: in our case 750x750 
down to 6000x6000 pixels) for the image pair and a dynamic-window search mode. Without any prior terrain 
knowledge, the method matches grid vertices at each pyramid level. The process encompasses four steps: 
1 ) first level (coarse) grid matching, 
2 ) second level grid matching, 
3) densification, 
4) intersection. 
The lack of a sufficient number of GCP’s (7), when this work has been done, and the very high number of grid 
matches (1666 for a match interval of 400m) in a wide overlapping area, made the selection of a small area around 
Thetis Bay mandatory (121 match locations generated with the same match interval). After the first level grid 
match, the visual inspection of the 121 match points yelded: 35 on water, 49 well correlated (within 1 pixel), 17 
within 5 pixels, the remaining 20 wrong (over 5 pixel). 
A subset of the first level match is given in Table 4. The quality measure (likewise a correlation 
coefficient) gives a rating out of twelve for each grid match (the higher the value the better). This is not always 
absolutely true: a value of 8.2693 (point 110), e.g., is not good because of a poor match due the snow fallen 
between the satellite acquisition of the images. The correction of 37 (17+20) grid vertices that didn’t correlate well 
in the right image were established manually obtaining, e.g., for the point 110 a quality measure of 6.2693 with a 
better match. Fig. 3 displays the grid vertices and their matches in a part of the study area after the first step, while 
Fig. 4 displays the grid vertices of the left image after the densification process which uses a 50m grid spacing; a 
mask was also introduced to eliminate points over the water area. The densification process doesn’t use the image 
pyramid model. The circles highlight areas with very poor contrast and different radiometric characteristics. Table 
5, a very small subset of the matches produced with the densification process, shows different kinds of matches. 
Intersection is the procedure of computing the three ground coordinates from the matched point pairs and 
the known satellite orientation (the “calculate X,Y,Z coordinate” box in the flow diagram); the related RMS error 
is treated as indicative. Two intersection steps were computed from the first level match (121 points) and from the 
densificated match (6561 points). With water areas having been masked, the former produced a mean RMS error 
of 17.727, while the latter a value of 1237.56 because of the large number of poor matches; in the last case, with 
1008 entries removed, the mean RMS lowered to 15.579. Table 6 is a subset of the values obtained in the former 
case. An offset of 65m was introduced to take into account the difference between ellipsoid (WGS84) and MHWM 
(Mean Height WaterMark). A comparison was also made with maps at 1:10000 scale: the 49 grid intersections 
with very good correlation were used in the successive step. 
Table 4. Quality Measure 
Image 
Point 
Pixel coord, x 
Pixel coord, y 
Quality measure 
10343 
109 
1456 
3704 
5.6257 
10342 
109 
2983 
3098 
5.6257 
10343 
110 
1496 
3704 
8.2693 
10342 
110 
2989 
3066 
8.2693 
10343 
111 
1096 
3744 
5.5708 
10342 
111 
2656 
3012 
5.5708