A ROBUST PHOTOGRAMMETRIC SYSTEM FOR WOUND MEASUREMENT
A. Malian *^, F.A. van den Heuvel*, A. Azizi ”
* Faculty of Civil Engineering and Geosciences, Delft University of Technology, The Netherlands
? Department of Geomatic Engineering, University of Tehran, Iran
a.malian@citg.tudelft.nl , F.A.vandenHeuvel@geo.tudelft.nl , aazizi@ut.ac.ir
Commission V, Working Group V/3
KEY WORDS : 3D Vision System, Feature Extraction, Robust Matching, Homomorphic Filtering, Wound Measurement
ABSTRACT
A MEdical Digital PHOtogrammetric System (MEDPHOS) is being developed that employs recent techniques in machine vision
and image processing to overcome major problems encountered in photogrammetric wound measurement. Conventional medical
wound measurement methods involve contact with the open wound, are uncomfortable and painful for the patient, carry a risk of
infection and are inaccurate. There is a demand to provide the medical society with a convenient, rapid, non-invasive and reliable
method for bedsore measurement. In this paper, a new strategy based on multiple-image feature-based matching based on the
Trifocal Geometry Constraint is presented. MEDPHOS is a trinocular vision system equipped with a projector. The projector serves
as a texture generator to compensate for the lack of natural object points on the wound. The method for finding corresponding points
in the three images requires calibration of the cameras to establish trifocal geometry. In order to improve the reliability of point
detection in the matching process a homomorphic filter is applied to reduce the effects of non-homogenous illumination as well as
specular reflectance caused by moisture on the wound surface. This process is followed by a morphologic TopHat operator and
connected component labelling for image segmentation and feature extraction. To eliminate the remaining ambiguities in
correspondences in case of a high density point field and to find the consistent triplets, auxiliary constraints are employed and a cost
function is minimised. It is shown that MEDPHOS is capable of reliably reconstructing 3D wound surfaces.
1. INTRODUCTION
Ulcers or pressure sores are chronic wounds of the skin
originate when immobile patients are exposed constantly to
high pressures on bony prominences. The clinicians need to
monitor the rate of healing by periodic measurement of any
changes which are brought about by therapy, the key elements
being surface area, volume of missing tissue and depth of the
wounds. Treating bedsore wounds places a large financial
burden upon the countries. According to the U.S. National
Decubitus Foundation (2001), the cost of bedsores in hospitals
is conservatively 55 billion dollars per year. Conventional
medical wound measurement techniques include acetate maps
for approximating area and alginate moulds or saline injection
for measuring volume. These methods involve contact with the
wound, are uncomfortable and painful for the patient, carry a
risk of infection and are inaccurate. Typically, the standard
deviation of area measurements is approximately 5% of the
wound area. In addition, different observers can show
considerable bias. An attempt has been made to reduce the
incidence of such errors by using active contour model to
improve the delineation (Jones, 1999). There is a demand to
provide the medical society with a convenient, rapid, non-
invasive and reliable method for bedsore measurement. Some
3D non-contact methods such as structured light and laser
scanning have been proposed (Plassman, et al., 1995). These
systems have some drawbacks that restrict their application. By
means of colour-coded structured light, for instance, the area of
skin ulcers and pressure sores can be measured with a precision
of about 5% provided that the area is greater than 9 cm? and is
at a distance of less than 3 cm. Volume can be estimated with a
precision of about 5% only if the volume-to-area ratio is greater
than 0.4 cm, i.e. the wounds are not shallow or small (Jones and
Plassman, 1995). Main disadvantages of these systems are that
they have low accuracy and are time consuming, expensive and
sensitive to motion of the patient. Close Range
Photogrammetry seems to be the best option for this kind of
applications. Research has been conducted to investigate the
capabilities of Least Square Matching (LSM) for modelling
wound surfaces (Boersma, et al, 2000). That method
sometimes fails due to differences in illumination, perspective,
reflectance as well as the lack of appropriate texture and hence,
no robust result could be found by LSM. The depth maps
produced by any area-based matching assumes a rich textured
area and that the observed scene is locally fronto-parallel which
causes problems for slanted surfaces (Alvarez, et al., 2002).
The delicate method of Geometrically Constrained Multi-Photo
Matching, is not appropriate for this application because it
needs accurate initial values (Shao, 1999). Moreover, the
Lambertian reflection rule does not apply for wounds and hence
the fundamental assumptions for an intensity-based matching is
no longer valid and the resulting object reconstructions are
prone not to be robust. The Lambertian rule states that the light
reflected by the surface of an object for a surface is the same in
all directions independent of the viewing angle. As a result, the
intensities at two corresponding points are equal. But, as an
ulcer is almost always very humid, there is a considerable
amount of specular reflectance on the wound surface leading to
many artefacts in non-homologous positions in the images.
Specular reflecting surfaces appear dark for points where the
reflection condition is not met and show specular reflexes for
the remaining points (Jahne and Hausecker, 2000). The
distinguishability of intensity-based methods is poor specially
for textureless objects, so these algorithms usually suffer from
slowness and ambiguous matches. Moreover, it should be kept
in mind that from a robustness point of view, adjustment by
least squares is weak, because of the smearing effect of the
gross observation errors which falsifies to considerable extend
the results of the adjustment (Ozanian, 1995). The intensity-
=264—
Ur " mn e fM MA M fco CD
Im es —
