PE&RS January 2016 - page 31

A Merging Solution for Close-Range DEMs to
Optimize Surface Coverage and
Measurement Resolution
Stephane Bertin, Heide Friedrich, and Patrice Delmas
Abstract
The process of efficient and effective DEM merging is increasing-
ly becoming more important. To allow DEM analysis for features
of different scales, an increase in surface coverage cannot result
in reduced measurement resolution. It is thus inevitable that
merging individual high-resolution DEMs will become common
practice for applications such as hydraulic roughness stud-
ies for fluvial surfaces. This paper presents an efficient and
effective merging solution, whereby accurate co-registration
of individual DEMs collected from consistent viewpoints and
standard averaging for overlapping elevations ensure seamless
merging. The presented method is suitable for DEMs collected
using any measurement technology, as long as individual DEMs
overlap and are arranged on regular grids. The merging solution
is applied to the study of a laboratory gravel bed measured with
vertical stereo photogrammetry at the grain scale (>106 points/
m2). We show that the approach can be integrated into the DEM
collection workflow at the design stage, which optimizes the
measurement performance. We present how resampling before
merging can be beneficial to keep data handling requirements
practical, whilst ensuring accurate surface representation.
Finally, the effect of scale variation is studied, showing that
seamless merging applies to DEMs with variable resolution.
Introduction
Problem Statement
Regularly gridded digital elevation models (DEM s) to sur-
vey areas of interest in both the laboratory and the field are
commonly used in fluvial hydraulics and the Earth sciences.
Often, a compromise has to be found between surface cov-
erage and measurement resolution. This is particularly true
when DEM s are obtained by photogrammetric means due to
the finite sensor size, and this also holds for other common
measurement techniques.
The trade-off between surface coverage and measurement
resolution in the photogrammetric application can be demon-
strated using the standard projective formulae, also com-
monly known as the pin-hole camera model. For a vertical
stereo-photogrammetric configuration between two identi-
cal cameras, the optical axes are parallel to each other and
perpendicular to the baseline. For such an arrangement, the
common field of view (CFoV) between the two images forming
a stereo pair is enlarged, whereby 3D information is extracted
by increasing the camera-to-object distance, Z:
CFoVW = −Zw
f
b (1)
CFoVH = Zh
f
(2)
where f is the camera focal length in pixel, b is the constant
baseline distance, w and h are the width and the height of
the cameras’ sensor in pixel, respectively, whilst the indices
W and H represent the directions parallel and perpendicular
to the baseline. Values are in metric units unless specifically
stated. The pixel size in the object space, Sp, which is the
minimum usable DEM grid size, and the theoretical depth
resolution, δZ, also called the minimum measurable depth, are
related to the camera-to-object distance as following:
S Zp
f p = (3)
δ Z = −
Z p
bf Zp
2
(4)
where p is the camera pixel size. One can see that by increas-
ing Z, both the horizontal (i.e., pixel size in the object space)
and the depth resolution will deteriorate.
Increasingly, studies’ regions of interest often exceed the
minimal measurement resolution required to record the sur-
face with solely one DEM. To address this issue, smaller DEM s
(thus of higher measurement resolution) can be recorded and
merged together to produce a DEM that has both, acceptable
surface coverage and measurement resolution. In the litera-
ture, DEM merging is also generally referred to as DEM mosaick-
ing, DEM stitching, and more generally data fusion. Examples
of applications in the Earth sciences can be found in the work
of Stojic et al. (1998), Butler et al. (2001), Chandler et al.
(2002 and 2001), Wackrow (2008) and Marzahn et al. (2012).
The Challenges
1. Visible seams can affect merged DEMs, degrading the mea-
surement accuracy and potentially impacting parameters
extracted from the DEMs (James et al., 2007). Those seams
are caused by horizontal and/or vertical shifts between
individual DEMs, where co-registration was insufficient.
2. The integration of the merging approach into the DEM
collection workflow requires two new parameters: (a)
the number of DEM s used for the merging, and (b) the
overlap between the DEMs. The two parameters are
Stephane Bertin and Heide Friedrich are with the Department
of Civil and Environmental Engineering, Faculty of Engineer-
ing, The University of Auckland, Private Bag 92019, Auck-
land 1142, New Zealand (s.bertin@ymail.com).
Patrice Delmas is with the Department of Computer Science,
The University of Auckland, Private Bag 92019, Auckland
1142, New Zealand.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 1, January 2016, pp. 31–40.
0099-1112/16/31–40
© 2015 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.83.1.31
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING January 2016 31