4.7 Height Precision and Accuracy
The heights obtained by automated matching are compared to the bundle heights. The average
and maximum differences between the two sets of heights are listed in Tables 1 and 2. The
average standard deviation of the heights from the matching algorithm and the root mean square
difference (RMSD) between the two sets,using filtering and no shaping parameters are listed in
Table 3.
AZ=05m AZ=125m
07 (cm) 14 1.2
cz (hg) 0.017 0.015
RMSD (cm) 2.7 4.1
RMSD (25. hg ) 0.033 0.050
Table 3. Precision and accuracy measures of the matching heights
The fit between the bundle heights and the heights from the matching is very good. However, it
must be considered that no "true" reference heights were available, so that the accuracy
measures mentioned here do always refer to the heights observed by an operator and adjusted
with a rather weak geometric configuration (one model with 6896 overlap), which can not be
objectively judged to be more accurate than the heights from the matching procedure.
The precision of the determined heights is very high. The results agree with other investigations
on precision of height measurements. Trinder in a recent paper reported on stereoscopic height
measurement precision ( Trinder 1986). For a wide angle camera, B/W film, 6 x magnification and
an overlap of 7096, he mentions a height precision of 0.037%. hg and an x-parallax clearance of
2.6 um independently of target size and shape. For a 19 x magnification - which was used for our
WILD AC1 measurements - he estimates respective values of 0.019%. hg and 1.4 um.
4.8 Reliability
Multiple solutions/side-maxima are strongly reduced with this matching technique, since the
patches are forced to move along the "relief" lines, following a certain step ratio. If we assume
that the left patch moves along its "relief" line with a step S, and the right with a step Sp (Figure
4), then the ratio S, / Sg is given with a very good approximation by
^
Si DXE + OYE = (18)
Sax +0
where DX,=àF,*/aZ , DY, =aF,Y/az (17a)
DXp=0Fp"/0Z , DYp=0F 7/2Z (17b)
- 29h -