the y-parallax after the orientation of 3.8 um corresponding to approx-
imately 0.15 of a pixel in the image. In a bundle adjustment the stan-
dard error of the matching was 3.4 um, which should be compared with
2.8 um using monocomparator measurements in the same investigation.
In an investigation concerning automatic parallax measurements in simu-
lated SPOT-images (1985) the author of this report came to the conclu-
sion that the obtainable two-dimensional matching accuracy with the
methods used, one correlation coefficient method and one least square
method with a window size of 17X17 pixels, was around 1/3 of a pixel
expressed as root mean square deviation. It was also concluded that the
error caused by radiometric differences between the images was larger
than the errors caused by the geometric differences between the images.
THE IMAGE MATERIAL
The images in the investigation were chosen as examples of practical
applications in photogrammetry. Large scale and small scale aerial
photographs and an example from close range photogrammetry (rock tunnel
wall). The following image data sets were used (the check points are
marked with crosses in the images).
Low. (above) An image pair in a scale of 1:4,000 of a suburban area
near Stockholm. The overlap in the strip is 607, the camera constant
152 mm and the base-to-height ratio 0.65. The photographs are of good
quality and manual measurements are easy to perform. The images are in-
sufficiently exposed in the shadows. Because of houses and other ob-
jects stereo shadows are common.
- 574 -