Boersma, Saskia M.
4 DIGITAL SURFACE MODEL GENERATION
4.1 The software for DSM generation
Softplotter, the program we used for the DSM generation, was developed for the use with aerial photographs. It
provides accurate and efficient extraction of terrain and feature data from digital 1magery. In the first step of the DSM
generation process the camera data is defined. We used the exterior orientations that are produced by the calibration
process. The images are corrected for the lensdistortions beforehand. This makes that only the focal length and the co-
ordinates of the fiducial marks had to be supplied. This information is different for each of the three cameras. The co-
ordinates of the artificial fiducials define the location of the principal point. An interior orientation had to be carried out.
No additional measurements in the images were necessary, because we used the exterior orientation results from the
calibration.
A. stereopair is created by resampling and rectifying the triangulated images into epipolar geometry such that the
y-parallax is removed and the x-parallax is interpreted as differences in elevation. The so-called surface tool is used to
create a Triangulated Irregular Network (TIN). This TIN collection is based on a least squares algorithm, which
establishes correspondences by minimizing the sum of the squared differences between the grey levels in patches of the
images. The strategy parameters are critical in producing a reliable TIN and affect the speed at which automatic
correlation is completed (Gooch, 1998). The autocorrelation procedure incorporates a hierarchical approach in which
correlations are performed at increasingly higher resolutions in the image pyramid. Interpolated points are ignored in
the creation process. Softplotter returns an indication of the accuracy of the created DSM.
4.2 Combining the three surface models
We combined the three models of the three possible combinations of two cameras to increase the reliability of the
created surface models. This is the advantage of using three cameras instead of the minimum of two cameras. With only
two cameras there is no check on the generated DSM.
Softplotter offers the possibility to compare two different DSMs. This creates the opportunity to assess the achieved
precision. The DSM that is generated in Sofplotter is a TIN. In order to be able to combine the three models, the points
have to be known in a regular grid. In Matlab, a software package specially designed for scientific and engineering
numerical computation, the points are interpolated to the same regular grid using bicubic interpolation. The sampling
interval is set to half the size of the grid size of the original TIN.
The three models are combined as follows. First the three TINs are interpolated to the same regular grid. Now we have
three heights for every point. The average height of two models is compared with the height of the third model. If the
difference is larger than twice the standard deviation computed from the differences between the two models, the value
of the third model is removed and a new DSM value is calculated. The resulting DSM is the average of the three
resulting models after removal of outliers. Optionally a 3 x 3 averaging filter is used to smooth the resulting DSM. This
final DSM can be used to calculate the wound features of interest, such as the volume, surface and contour.
4.3 Determination of the volume
In order to determine the volume of a wound two surfaces have to be known: the measured surface of the wound and
the original, healthy skin surface. The volume of the wound is sandwiched between those two surfaces. The original
healthy skin has to be reconstructed from undamaged parts of the skin surrounding the wound. As a first step, the
outline of the wound has to be defined. The second step is the reconstruction process itself. An interpolation method,
such as cubic splines, has to be used to interpolate the original healthy skin (de Jong, 1997). As a final step, the area,
volume, circumference and depth of the wound can be calculated from a three-dimensional grid created by the process
descried above. It has to be stressed that the majority of the errors in the reconstruction of the healthy skin will be
eliminated in the difference between two volumes. These volume changes are used for assessing the healing process.
Neglecting the effects of errors in the reconstruction of the healthy skin, the precision of the volume depends on the
precision of the generated DSM and the depth of the wound. The precision of the volume can be approximated with:
ep!
= vol Oh
Cot = VOI -- à - Surf - O yop
dept
When the accuracy of the determined heights is 0.5 mm, and the wound has a depth of 5 cm, the accuracy of the volume
equals 1% of the total volume.
88 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.
——
44
Differe
experin
generat
lack of
experin
project
Figure
resultin
The the
and an
This re:
are Cor
summa
models
distribu
Object
Plastic
Head
Pressur
Pressur
The tal
height.
better f
pressut