errors in the indications of the machine. On the other hand, the machine 
provides a very reliable, and reasonably convenient three-dimensional test 
field for calibrating photogrammetric systems. We were fortunate to have 
contact with the Quality Assurance Directorate (Ordnance) of the Ministry of 
Defence at the Royal Arsenal Woolwich, in whose establishment is an 
exceptionally well designed three-axis machine produced by the Cranfield Unit 
for Precision Engineering (C.U.P.E.). This machine has a design volumetric 
accuracy of 50pm, with an available travel of 4m X 2m X 0.76m, and is shown 
in plate 1. The wide angle laser source was mounted on the quill, and the two 
UMK 10FP cameras set up on the machine table, as shown. The plates used were 
Agfa 10E75, which have a very fine grain (less than 1pm), very slow speed 
(less than 1ASA) and very high contrast emulsion. This ensured that the 
recorded images were as sharp, and as well defined as possible. The lens 
shutters were opened permanently, and exposure (1sec.) controlled by a paper 
lens cap. Exposures were made, on the same plate, of the probe at 48 points 
regularly spaced within a 1.6m X 1.5m X 0.6m volume. The resulting 
photographs were clear plates with 48 sharp elliptical images. The plates 
were measured ( with a repeatability of 2um) on a Zeiss Jena ZKM 01 250 D 
measuring microscope, and the results processed using Granshaw's bundle 
solution. 
It is well known that at least three photographs are necessary for 
self-calibration, so that this could not be used.A form of calibration was 
achieved, however, by using a third of the points as "knowns" (ie. machine 
Xyz co-ordinates accepted as error free). The resulting calibration 
parameters were then held fixed in the bundle solution, and the remaining 
points co-ordinated. The standard deviation of the photograph co-ordinates 
was found to be 2.8)m, which corresponds to a standard deviation of 70pm 
against the machine co-ordinates. This is very much better than the original 
"cube" test at University College (Table 1). 
Measurement of records 
In the attempts described above to extend classical practices of close 
range photogrammetry, some improvements have been made to the process of 
measuring the records. Monocomparator measuring was used initially, but most 
of the later plates have been measured visually on a Zeiss Jena ZKM 01 250D 
two co-ordinate measuring microscope, equipped with Heidenhain moiré scales 
reading to 0.11pm, Recently, provision has been made for automatic recording 
of the readings on punched tape for direct input to the computation of the 
photogrammetric solution. This will speed up the process, and will reduce the 
risk of reading errors, and will justify further attention to the control of 
the optical measurement process, and to the possibility of partial 
automation. Specialized sensor techniques already being developed offer the 
possibility of replacing the eye as the critical factor in the setting on the 
record of each point. Parallel developments in the optical systems employed 
in cameras will demand specialized sensors in any case, but from the evidence 
presented here it may be deduced that there is already a promising 
opportunity to begin to apply automatic setting and measurement to the 
Specialized recordings we are obtaining from conventional cameras. 
Effects of photographic plate flatnesss in photogrammetry 
Conventional photogrammetry assumes that the image is formed on a 
perfectly flat photographic plate, which is subsequently measured on an X-y 
measuring machine or plotter. If the plate is not flat, then errors in 
position will occur in regions of the plate which are not in the intended 
190