(Our response to Chandler et al.)
A paper recently published in Mathematical Geology raises some
concerns, which need to be expressed. The paper is entitled "Hillslope
topography from unconstrained photographs" by Arjun Heimsath and Hany
Farid, and presents a "simple methodology to extract the high
resolution 3-dimensional topographic surface from photographs taken
with a hand-held camera with no constraints imposed on the camera
positions or field survey".
The implication from the abstract is that extracting spatial data from
ground-based photography is a new development. Unfortunately, this
prompted a BBC reporter here in the UK to publish an article on the
web (Anon, 2002), apparently based on a web news item from Heimsath
and Farid's own institution, which did not even mention
photogrammetry, the science associated with extracting spatial data
from photographs that has evolved over 150 years! During the same
period these techniques have spawned an industry that has provided
most of the world's large scale mapping and, as websites of the Remote
Sensing and Photogrammetry Society's (www.rspsoc.org) and the American
Society of Photogrammetry and Remote Sensing (www.asprs.org) suggest,
so much more!
In fairness to the authors, the science of photogrammetry is referred
to and relevant work involving photogrammetric methods is acknowledged
in the paper, which was available freely over the Internet. However,
from the Introduction and Discussion it would seem that the authors
are convinced that photogrammetry imposes "high degrees of constraints
upon the positions of the cameras", and that "precisely located
control points" and "application specific third-party software" are
required, and they go on to argue that their solution is an
improvement.
Firstly, it is important to state that in modern photogrammetry there
is huge flexibility in both the position and orientation of the camera
and indeed the type of camera that can be used. The key condition is
that in order to extract 3D coordinates, the feature required must
appear on a minimum of two photographs. It is important to maintain
some general geometric relationships between the camera positions and
the object, which are embodied in the base/height ratio (see any
textbook on photogrammetry, Slama, 1980, p. 952; Wolf, 1974, p. 66;
Mikhail, Bethel, and McGlone, 2001, p.28). This is also an issue that
would be of some importance to the approach developed in the paper,
although sadly this aspect is not examined. It is also not essential
to use precisely located photo-control points to carry out
photogrammetry, although there are many advantages from doing so. Most
significantly for process-based studies, photo-control defines an
appropriate datum in which elevation changes are related to the local
vertical and allows meaningful spatial comparison in multi-epoch
surveys.
There are also some problems with the theoretical aspects of the
development. The authors assume a perfect pinhole camera as do
photogrammetrists. From this assumption the well- known and
fundamental collinearity equations are developed. However, the authors
do not consider the impact of camera geometry and lens distortion in
their mathematical model, which is important for measurements derived
from any camera not designed for photogrammetry. This is an area
where photogrammetrists have made major contributions (Kenefick, Gyer,
and Harp, 1972, p 1118; Slama, 1980, p. 480; Fraser, 1997), to the
extent that the mathematical models used to compensate for a whole
array of systematic effects associated with modern sensors have become
broadly accepted. Such models and appropriate photo-control points can
assist in the camera calibration process, essential if 35mm imagery is
used. Another key limitation of the new approach is the whole
principle of the "paraperspective" projection. Why use a non-rigorous
model when a rigorous solution is clearly in the public domain? (i.e.
collinearity). The "paraperspective" model may be appropriate for the
machine vision community where camera to object distances are
typically less than a few metres, but at the process-based scale this
model is surely not accurate enough, particularly if there is a wide
depth of interest in the object space. This limitation is clearly
responsible for the "subduing" effect on the measured topography and
for the mean elevation errors, which most photogrammetrists would
consider to be rather large considering the photo scale (again not
mentioned).
The authors make the point that photogrammetry applied to past
landform studies has relied upon third-party software. Why is this
such a limitation? Did the authors write their whole paper in a basic
ASCII editor such as Microsoft "notepad" or did they use a more
comprehensive word processing package such as Microsoft Word? The cost
of proprietary photogrammetric software packages capable of extracting
DEMs consisting of many thousands of points automatically and then
generating orthophotographs has never been cheaper. They are often
directly available to university researchers through generous academic
licensing agreements and generally easy to use. If direct access to
such functionality proves difficult, why not contact and collaborate
with researchers engaged with photogrammetric research? From our
experience, this can be a "win-win" situation for all concerned!
In summary, it is somewhat surprising that the referees selected to
assess, review and improve this particular paper have not identified
these weaknesses on this occasion. Of perhaps greater concern is that
the article may encourage geographers and earth scientists to use the
techniques offered freely and which many would classify as being
photogrammetric. Poor results will inevitably be obtained and this may
discourage a whole new generation from benefiting from 150 years of
scientific evolution in photogrammetry.
J.H. Chandler, Senior Lecturer in Photogrammetry and Land Surveying,
Loughborough University and Vice-Chair- Remote Sensing and
Photogrammetry Society, UK.
S. Robson, Lecturer, Photogrammetry with Remote Sensing, University
College London, UK.
J.P. Mills, Lecturer in Geomatics, University of Newcastle upon Tyne,
UK.
S.N. Lane, Professor in Physical Geography, University of Leeds, UK.
K. B. Atkinson, Editor Emeritus, Photogrammetric Record, UK.
P.R.T. Newby, Editor, Photogrammetric Record, UK.
F. Varkaris, GIS Application Engineer, Keigan Systems Inc., Canada
References