![](bg1.png)
Continuous Tracking of Structures from an Image Sequence
Yann Lepoittevin
1,2
, Dominique B
´
er
´
eziat
3
, Isabelle Herlin
1,2
and Nicolas Mercier
1,2
1
Inria, B.P. 105, 78153 Le Chesnay, France
2
CEREA, Joint Laboratory ENPC - EDF R&D, Universit
´
e Paris-Est,
Cit
´
e Descartes Champs-sur-Marne, 77455 Marne la Vall
´
ee Cedex 2, France
3
Universit
´
e Pierre et Marie Curie, 4 place Jussieu, Paris 750005, France
Keywords:
Tracking, Motion, Data Assimilation, Satellite image, Meteorology.
Abstract:
The paper describes an innovative approach to estimate velocity on an image sequence and simultaneously
segment and track a given structure. It relies on the underlying dynamics’ equations of the studied physical
system. A data assimilation method is applied to solve evolution equations of image brightness, those of
motion’s dynamics, and those of distance map modelling the tracked structures. Results are first quantified
on synthetic data with comparison to ground-truth. Then, the method is applied on meteorological satellite
acquisitions of a tropical cloud, in order to track this structure on the sequence. The outputs of the approach
are the continuous estimation of both motion and structure’s boundary. The main advantage is that the method
only relies on image data and on a rough segmentation of the structure at initial date.
1 INTRODUCTION
The issue of detecting and tracking a structure covers
a broad of major computer vision problems. Read-
ers can refer to (Yilmaz et al., 2006), for instance,
in order to get an extensive description on this issue.
However, images may be noisy, as this is the case for
satellite acquisitions, and assumptions on dynamics
should then be involved. To our knowledge, no paper
concerns a method that simultaneously estimates mo-
tion and segments/tracks a structure from only image
data and a rough segmentation of the structure. How-
ever, methods exist that segment and track a structure,
given motion field and initial segmentation (Peterfre-
und, 1999; Rathi et al., 2007; Avenel et al., 2009), or
that track a structure and estimate its motion if this
structure has been accurately segmented on the first
image (Bertalm
´
ıo et al., 2000).
The use of data assimilation recently emerged in
the image processing community. In (B
´
er
´
eziat and
Herlin, 2011), motion estimation is discussed, and
solutions are described for processing noisy images.
In (Papadakis and M
´
emin, 2008), an incremental 4D-
Var is used, that also computes motion field and tracks
a structure, but relies, as inputs, on both image data
and accurate segmentation of the structure on the
whole sequence. Our approach has the advantage to
simultaneously solve the issues of motion estimation,
detection, segmentation and tracking of the structure,
based, as only inputs, on image data and their gradient
values.
Section 2 describes the main mathematical com-
ponents of the approach. Sections 3 and 4 discuss
results obtained on synthetic data and meteorological
satellite acquisitions. Section 5 concludes with some
remarks and perspectives on the research work.
2 MATHEMATICAL SETTING
Our approach is based on a 4D-Var data assimilation
algorithm, used to estimate motion on the sequence
and track a structure.
Ω denotes the bounded image domain, on which
pixels x =
x y
T
are considered, [0, T ] the studied
temporal interval, and A = Ω × [0, T ].
2.1 Model of Structure and Input Data
Let define the structure tracked along the image se-
quence by an implicit function φ (see Figure 1): each
pixel x at date t gets for value its signed distance to
the current position of the structure boundary.
Observations, used during the assimilation, are
images themselves and their contour points, obtained
by thresholding the maxima of the gradient norm.
386
Lepoittevin Y., Béréziat D., Herlin I. and Mercier N..
Continuous Tracking of Structures from an Image Sequence.
DOI: 10.5220/0004278503860389
In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2013), pages 386-389
ISBN: 978-989-8565-48-8
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)