International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 5. Hakodate 1998
Extraction of Moving Ships from Navigational Images
Akio Satori, Mitsuo Suzuki, and Masaki Oshima
Department of Electronic and Mechanical Engineering
Tokyo University of Mercantile Marine
2-1-6, Ecchujima, Koto-ku, Tokyo, 135-8355
E-mail:oshima@ipc.tosho-u.ac.jp
JAPAN
Commission V, Working Group IC V/III
KEY WORDS: dynamic analysis, optical flow, template matching, navigation
ABSTRACT
Experiments to extract and track ships in the spatiotemporal images are performed aiming to provide useful
supporting information for ship maneuvering instead of human eyes in the watching. Two computer vision
techniques are applied to this purpose. The first is the template matching method and the second is optical
flow method. These methods are applied complimentary and extracted information are integrated. The results
for a set of sequential images on the sightseeing ship have been so far satisfactory.
1. Introduction
In maneuvering ship on the ocean, judging the sur-
rounding situation properly is very important to se-
cure the safety of the ship. To provide useful support-
ing information for ship maneuvering various systems
such as ARPA system have been developed. However,
to watch circumstances by operators is still important
particularly under severe conditions such as navigation
in passing narrow channel. It is desirable to provide
new system in addition to the previously developed
supporting systems of ship handling, because the con-
fidence of the total system will be improved by plural
different techniques. In this background, we make ex-
periments to extract and track ships in the spatiotem-
poral images. There have been some researches to ex-
tract ships in the image. Typical approaches [1], [2]
have been to extract a ship in a still image by vision
technique. This means tracking of a ship is not accom-
plished. Our approach is to extract and track ships
in the spatiotemporal images. Two computer vision
techniques are applied to this purpose. The first is
the template matching method and the second is op-
tical flow method. These methods are applied compli-
mentary and extracted information are integrated. In
the following sections, outline of the experiments is de-
scribed in section 2, template matching and tracking
are described in sections 3. The optical flow method is
described in section 4. Experimental results are shown
in section 5. Section 6 concludes our approach.
Figure 1: Image of a ship
2. Outline of the experiments
The problem to be solved here is to detect and track
a moving ship in a field of view from a set of sequential
images. The figure 1 shows one of our target scenes. To
extract moving objects, some approaches are known in
vision techniques. If an object shape (call it template)
is known previously, it is able to extract it from each of
sequential images [3]. A set of detected positions can
provide tracking information. But, reliable extraction
of a ship in an image is still an open problem if the
sit
ent si
shape
prep?
a difl
adop!
infor!
ship
for d
wake
ator
part.
tatio:
meth
tract
to ol
tive
the s
the c
ified
tend
to b
tical
used
the
pixel
pare
devi:
opti
and
for
prog
plat
flow
in t]
a Ve
is jt
The
plat
The
plat
n-tl
foll
bin.
cal
of t
(cal
wid
to 4
pro
