4.3 How Well Does the 
Criteria? 
Instrument Meet Design 
A) Time to record: Cameras are easily timed to 
record in less than 0.1 seconds. This criterion is 
easily satisfied with the conventional two camera 
approach. 
B) Output of results: Results are presently 
displayed using contours. Digital surface model 
matching has not yet been incorporated into the 
software. 
C) Reliability: The reliability of matching obtained 
in the test was surprisingly high by both approaches, 
largely due to the robustness of the least squares 
matching procedure, using the selected run-time 
parameters. If the method fails it will usually be 
only in the vicinity of areas in the vicinity of 
hair, eyebrows, eyes, and so on, which are not 
relevant to the practitioner nor to surface model 
matching. It may fail in the vicinity of steep 
slopes around the nose, or at eyebrows or eyes, but 
again what is relevant to the practitioner is the 
soft tissue around the cheeks, chin and forehead and 
as such disruptions to the smooth skin and steep 
slopes need not be a problem, especially if digital 
surface matching is to be used. A surface measuring 
system must be capable of coping with different skin 
types and colours, a matter which remains 
undetermined in this case. 
D) Convenience: The operation has so far been fully 
automatic in all modes, owing largely to the 
selection of parameters appropriate to the surface in 
question. The instrument will operate in a normally 
lit room. 
E) Accuracy: Current accuracy as estimated from the 
tests as described above is adequate for the 
particular uses expected to generate significant 
demand. 
F) Resolution: The levels of resolution which have 
been achieved - see Table 1 - are adequate for the 
anticipated uses. 
G) Cost: The hardware required for the current 
instrumentation comprises the cameras, the 
illumination and texturisation apparatus (specially 
fabricated), a typically-configured PC with VGA 
screen for image display, frame-grabbers boards and 
a separate monitor for the cameras. Proprietary 
software is used for surface contouring and display. 
H) Computation Speed: Considerable programming 
effort has been devoted to ensuring a short and 
probably acceptable computation time, although not 
strictly real time. 
ani = 
Wl E. 
a 
  
Figure 2: 1 mm contours of part of abdomen, based 
on 421 points, as determined using symmetry 
detection followed by least squares matching. 
I) Safety: The procedure is safe, and moreover, 
using digital surface model matching avoids the need 
for control points or fixing the patient with respect 
to cephelostats, for example, and makes the method 
comfortable for the patient. 
This instrument may be compared to recent similar, 
digital photogrammetric approaches to bio-medical 
measurement, e.g. by Grün & Baltsavias (1988), 
Trinder et ai. (1990), and Deacon et al. 
(1991), and to active triangulation approaches, e.g. 
Frobin & Hierholzer (1991). However, these 
approaches all differ in the way they meet the 
criteria mentioned above, and the task of the 
comparison becomes difficult and is perhaps not 
constructive. As no body surface measuring devices 
are used in the health services of the local region 
at present, it seems reasonable to foresee use by the 
medical practitioners with whom the writer has been 
in contact. 
  
Figure 3: 1 mm contours of a human face, derived 
from 736 points. 
5. CONCLUSIONS 
The paper provides a brief case-study of a device 
which is capable of determining bio-medical soft 
tissue surface shapes, which operates reliably and 
without operator intervention, and which has been 
brought to operating level. A feature of the system 
may be its simplicity: it has been surprising what 
could be achieved with a ordinary PC, although with 
extensive programming effort in image processing and 
with carefully chosen run-time parameters. The 
difficult facets of the development have not been 
seen as achieving precision from the least squares 
matching but achieving reliability and an acceptable 
speed. 
The projected texturising grid pattern facilitates 
target searching, and least squares image matching 
does not suffer. However, target detection can be a 
slow procedure because of the volume of image 
processing and judicious selection of an appropriate 
strategy can be crucial. Different levels of 
resolution have been found possible: speed can be 
achieved at the expense of resolution. Low 
resolution is seen to be suited to a number of 
medical applications, especially as it provides a 
surface model little different from the high 
resolution models for limbs and backs. 
Hardware and software development will continue, to 
advance accuracy, reliability, resolution, coverage 
and speed. All hardware components must be 
thoroughly calibrated. Digital surface model 
matching must become an integral part of the device. 
   
    
CN > OO HI rrj CC Mi eed A QUI EL. LX au S Vi QN EX ees tO MUN NR 
Iri