PE&RS January 2018 Full - page 33

Detecting Glacier Surface Motion by Optical Flow
M.G. Lenzano, E. Lannutti, C. Toth, A. Rivera, and L. Lenzano
Abstract
In this study, we assessed the feasibility of using optical flow,
in particular, large displacement optical flow (
LDOF
) method
as a possible solution to obtain surface movement data to
derive ice flow velocities in a glacier. Tests were carried
out at the Viedma Glacier, located at the South Patagonia
Icefield, Argentina, where terrestrial monoscopic image
sequences were acquired by a calibrated camera from April
2014 until April 2016. As for preprocessing, the Correlated
Analysis method was implemented to avoid and minimize
errors due to the measurable changes in lighting, shad-
ows, clouds, and snow. The results show a flow field with a
maximum surface velocity value of 3.5 m/d. The errors were
minimized by averaging the image sequence results based
on seasons, in which the Total Error Reconstruction yielded
fairly good mean accuracy (0.36 m/d). In summary, it was
demonstrated that
LDOF
can provide accurate and robust
solution to detect daily changes in the glacier surface.
Introduction
Geospatial data acquisition methods for Earth observation and
monitoring applications have seen great technological advance-
ments in recent years (Bartholomé and Belward, 2005), as
the performance potential of the sensors, in terms of spatial,
spectral, and temporal resolutions has significantly expanded.
Consequently, nowadays remote sensing techniques represent
an attractive and affordable approach to study different natural
phenomena, such as glaciers, where the main advantage is a
simpler and more economical implementation compared to
conventional field surveying measurements. At present, optical
imagery is the most commonly used sensory data to monitor gla-
ciers because it is an efficient low cost method, used since mid-
1980s for mapping surface velocities. (Heid and Kääb, 2012).
To study glaciers, a variety of methods and techniques
have been used since the mid-nineteenth century (Gao and
Liu, 2001; Bamber and Rivera, 2007); typically a combina-
tion of these is required to obtain high resolution spatial and
temporal observations needed to model the complexity of the
dynamics processes. Airborne/spaceborne remote sensing
easily provides consistently accurate low and medium resolu-
tion data over large areas (Toth and Jóźków, 2016). In contrast,
terrestrial or close-range sensors (Moustafa, 2000) can provide
high resolution observations (Schwalbe
et al
., 2016), and
thus difficult to map topographies, such as glaciers, can be
surveyed with high accuracy for smaller areas. Satellite-based
optical sensors may often have limiting factors for observa-
tions, such as cloud cover, especially in mountain regions
(Gleitsmann and Kappas, 2006). Since glaciers are dynamic
objects, the temporal aspects of the mapping are equally
important, and therefore, the revisit time of satellite plat-
forms may present some disadvantages. A permanent sensor
installation could offer unprecedented temporal resolution
and observation capabilities (Toth and Jóźków, 2016), and us-
ing time-lapse imagery provides a viable approach to glacier
change detection (Harrison, 1992; Hashimoto
et al
., 2009; Ahn
and Box, 2010; Maas
et al
, 2010; Rivera
et al
., 2012b; Daniel-
son and Sharp, 2013; Lenzano
et al
., 2014).
The dynamics of glaciers has noticeably changed due to
global warming during in the late 20
th
and 21
st
centuries (Oer-
lemans, 2005; Bolch
et al
., 2012), and thus, to properly assess
and monitor their dynamics requires the detailed mapping
of surface velocities and changes in geometry (Howat
et al
.,
2007); noted that subsurface velocities are also of importance,
but they cannot be easily observed. Ice flow velocities vary
along glaciers, following complex patterns defined by stress
and strain rate distributions (Benn and Evans, 2010). In order
to map glacier surface velocities, various methods have been
used, including traditional point-based surveys, and then using
remote sensing methods that can provide mass points in irregu-
lar or regular grid distributions. Photogrammetric and comput-
er vision methods are available to obtain a dense and accurate
grid adequate to support glacier dynamic analysis (Matías
et
al
., 2009; Westoby
et al
., 2012; Ryan
et al
., 2015, Piermattei
et
al
., 2015), and may provide various metric products, such as
ice velocities, volume changes, etc. Note that photogrammetry
is fundamentally concerned about the metric integrity of the
derived products, while computer vision is primarily focused
on the correct recognition and reconstruction of objects.
Clearly, photogrammetry and computer vision share, or at least
should share, a common (or at least widely overlapping) theo-
retical basis (Granshaw and Fraser, 2015). Certainly, exploiting
the complementarities of the two disciplines may provide ac-
curate solutions to monitoring many natural phenomena.
Image sequence processing in computer vision is a broad
field, and includes tracking, structure from motion (
SFM
),
optical flow, etc. These subfields are closely related, and
the concept is that the object/observer movement enables to
obtain accurate information of the sensed changes over time,
which is mostly based on identifying conjugate geometrical
primitives in the images. Finding correspondence between
pairs of points in two images remains one of the fundamental
computational challenges in computer vision (Wedel
et al
.,
2009). At the beginning of the 1980’s, techniques based on
coarse-to-fine strategies were developed, such as intensity-
based optical flow algorithms, and then quantitative methods
of computer vision, including many early feature-based stereo
correspondence algorithms (Szelisky, 2010). In some cases,
different motion and structure paradigms were developed,
using optical flow as an intermediate representation of mo-
tion correspondences between image features, correlations,
or properties of intensity structures (Beauchemin and Barron,
M.G. Lenzano, E. Lannutti, and L. Lenzano are with
the Departamento de Geomática. Instituto Argentino
de Nivología, Glaciología y Ciencias Ambientales-CCT,
CONICET, Mendoza, Argentina
(
).
C. Toth is with the Department of Civil, Environmental and
Geodetic Engineering, The Ohio State University, Ohio, USA.
470 Hitchcock Hall, 2070 Neil Ave., Columbus, OH 43210
A. Rivera is with Casilla de correo 300, Av. Ruiz Leal s/n.
Parque Gral. San Martín. Mendoza. Argentina. CP: 5500.
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 1, January 2018, pp. 33–42.
0099-1112/17/33–42
© 2017 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.1.33
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
January 2018
33
I...,23,24,25,26,27,28,29,30,31,32 34,35,36,37,38,39,40,41,42,43,...54
Powered by FlippingBook