1995). The optical flow extracted from imagery is the result
of the apparent movement pattern between objects, caused
by either relative deformation or absolute movements. The
objective of motion estimation is to compute an independent
estimate of motion for each pixel, which is generally known
as optical flow (Szelisky, 2010). Optical flow-based methods
can be classified as differential, correlation, frequency, and
variational (De la Nuez, 2010). The variational methods have
demonstrated the best performance, allowing the accurate es-
timation of dense flow field based on the original formulation,
introduced by Horn and Schunck (1981).
In spite of the existence of a variety of optical flow tech-
niques, the majority of the algorithms concern small displace-
ments and only a few procedures have been developed to
detect large displacements, such as work by Weinzaepfel
et al
.
(2013). With respect to the image acquisition rate, changes oc-
curring in time-lapse images of glaciers, where for half of the
day no imagery can be acquired, the changing displacement
could be large. The Large Displacement Optical Flow (
LDOF
)
method (Brox
et al
., 2004; Brox
et al
., 2009; Brox and Malik,
2011) offers a powerful solution to estimate large displace-
ments in image sequences, and is based on a solid numerical
method that combines descriptor matching with the variation-
al model, and uses a coarse-to-fine strategy with the so-called
warping technique. Finally, the descriptor matching and the
discrete optimizations can provide sub-pixel accuracy.
Only a few studies of ice motion have been carried out us-
ing optical flow algorithm, such as using satellite and terrestri-
al images by (Vogel
et al
., 2012) and (Bown, 2015), respectively.
In this work, we propose to use the
LDOF
algorithm to estimate
the glacier motion based on terrestrial, monoscopic time-lapse
image series, acquired by non-metric professional
DSLR
cameras
systems. The optical flow method from computer vision will
provide the change detection of 3D objects, and the photogram-
metric processes will supply the scaling and the metric charac-
terization of the glacier movement. This study aims to obtain
accurate solutions at pixel level with high temporal and spatial
resolution, and determine ice velocities of the glacier termi-
nus. The test was carried out at the Viedma glacier, Southern
Patagonia Icefield (
SPI
), Argentina between 2014 and 2016. This
site is an important calving glacier in the region covering an
area of 945 km
2
(Aniya
et al
., 1996). Since the behavior of the
glacier at the terminus is of high interest, this area was selected
for the investigation. The investigation is a continuation of our
previous effort (Lannutti
et al
., 2016; Toth
et al
., 2016). The
outline of this paper is as follows: the next Section provides
a detailed description of the study area and data collection,
followed by a review of the methodology proposed. Then, the
results with analysis is provided leading to the conclusions.
Test Area and Data Acquisition
The South Patagonia Ice field (
SPI
) is located in South Ameri-
ca, Argentina and Chile, covering an area of 13,000 km
2
with
an average length of approximately 30 to 40 km at a mean
altitude of 1,191 m
ASL
(Aniya, 2013). Viedma glacier is lo-
cated at 49° 31
′
S, 72° 59
′
W, Los Glaciares National Park,
SPI
,
Santa Cruz, Argentina, see Figure 1. The glacier was selected
for this study due to the availability of earlier investigations,
carried out over the past 30 years (Skvarca
et al
., 1995; Aniya
et al
., 1996; Lopez
et al
., 2010; Riveros
et al
., 2013), and its
representativeness at the
SPI
.
To support the field image acquisition of the time-lapse
imagery, an integrated data acquisition system was built
based on a CANON EOS Mark II
DSLR
camera (C1); pixel size:
7.2 μm, objective focal length: 50 mm, and FOV: 46°. The C1
was calibrated prior to field deployment by the United States
Geological Survey (
USGS
). Also, a NIKON D3 camera (C2) cali-
brated by Rollei Metric was used for preprocessing purposes;
pixel size: 8.5 μm, objective focal length: 35 mm, and FOV:
62°. Both systems are powered by one 12V/7Ah lead acid bat-
tery, charged by two 38W solar panels. The cameras with the
supporting electronic systems are protected by a waterproof
box with a visor to reduce reflections. An inspection port in
the rear of the enclosure provides visual access for monitor-
ing the camera operation. The image acquisition systems were
installed on a rigid metal structure, fixed to outcrops of the
south margin of the Viedma Glacier. The C1 was located 70 m
Figure 1. Map of the study area with each camera location and
FOV
.
34
January 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
