Optik
124 (2013) 5357–
5362
Contents
lists
available
at
ScienceDirect
Optik
jou
rn
al
homepage:
www.elsevier.de/ijleo
An
improved
building
boundary
extraction
algorithm
based
on
fusion
of
optical
imagery
and
LIDAR
data
Yong
Li
a,∗
,
Huayi
Wu
b
,
Ru
An
a
,
Hanwei
Xu
a
,
Qisheng
He
a
,
Jia
Xu
a
a
School
of
Earth
Sciences
and
Engineering,
Hohai
University,
Nanjing
210098,
China
b
State
Key
Laboratory
of
Information
Engineering
in
Surveying,
Mapping,
and
Remote
Sensing,
Wuhan
University,
Wuhan
430079,
China
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
12
September
2012
Accepted
20
March
2013
Keywords:
LIDAR
Optical
image
Fusion
Boundary
extraction
Building
Mathematical
morphology
a
b
s
t
r
a
c
t
This
article
presents
a
new
method
of
automatic
boundary
extraction
using
LIDAR-optical
fusion
suited
to
handle
diverse
building
shapes.
This
method
makes
full
use
of
the
complementary
advantages
of
LIDAR
data
and
optical
imagery.
Different
building
features
are
extracted
from
the
two
data
sources
and
fused
to
form
the
final
complete
building
boundaries.
First,
the
points
of
each
roof
patch
are
detected
from
LIDAR
point
cloud.
This
process
consists
of
four
steps:
filtering,
building
detection,
wall
point
removal
and
roof
patch
detection.
Second,
initial
building
edges
are
extracted
from
optical
imagery
using
an
improved
Canny
detector
constrained
by
edge
location
information
derived
from
the
LIDAR
point
cloud
as
edge
buffer
areas.
Finally,
the
roof
patch
and
initial
edges
are
integrated
by
mathematical
morphology
to
form
the
final
complete
building
boundaries.
All
processes
have
no
constraints
or
rules
on
building
shapes.
This
method
is
fully
data-driven
and
suitable
for
any
building
shape.
LIDAR
data
and
aerial
images
of
complex
geographical
environments
are
used
to
test
the
method.
These
experimental
results
demonstrate
that
our
method
can
automatically
extract
accurate
boundaries
for
buildings
with
complex
shapes,
and
also
is
highly
robust
in
complex
environments.
© 2013 Elsevier GmbH. All rights reserved.
1.
Introduction
Accurate
building
boundaries
are
important
for
diverse
applica-
tions
such
as
real
estate,
city
planning,
and
disaster
management.
The
automatic
extraction
of
building
boundaries
is
challenging
due
to
building
shape
variability
and
surrounding
environmental
com-
plexity.
High-resolution
optical
imagery
contains
rich
spectral
and
textural
information
and
is
easily
affected
by
many
factors
such
as
contrast,
illumination
and
occlusion.
The
airborne
Light
Detection
and
Ranging
(LIDAR)
can
rapidly
acquire
dense
and
precise
height
data
of
large-scale
areas
by
emitting
and
receiving
laser
pulses.
The
height
changes
are
more
suitable
for
locating
building
boundaries
more
than
the
spectral
and
texture
changes.
However,
the
hori-
zontal
accuracy
of
boundaries
extracted
from
LIDAR
data
is
poor
because
of
laser
pulse
discontinuousness.
Hence
a
number
of
methods
have
been
developed
to
make
use
of
LIDAR
point
clouds
to
extract
building
boundaries
and
solve
the
horizontal
accuracy
problem
arising
from
laser
footprint
dis-
continuousness.
These
approaches
are
broadly
divided
into
three
categories.
The
first
category
is
simplifying
and
generalizing
the
coarse
edges
derived
from
raw
data
based
on
some
building
shape
hypotheses
such
as
parallel
or
perpendicular
edges
[1–7].
However
∗
Corresponding
author.
Tel.:
+86
13655192670.
E-mail
address:
liyong@hhu.edu.cn
(Y.
Li).
it
is
difficult
to
judge
whether
the
hypotheses
are
in
conformity
with
reality.
The
second
category
obtains
complete
building
bound-
aries
by
grouping
the
extracted
features
such
as
corner
points,
line
segments
and
roof
planes
[8–16].
The
feature
grouping
pro-
cess
follows
complex
rules
which
are
crucial
for
the
success
and
robustness
of
this
method.
This
category
of
methods
performs
well
mainly
on
the
regular
buildings
composed
of
straight
lines
[17,18]
.
The
third
category
uses
model
primitives
to
determine
building
boundaries
by
fitting
input
data
to
the
adopted
mod-
els
[19,20].
The
adopted
model
primitives
are
often
some
regular
shape
or
a
designed
model
dataset.
It
is
difficult
to
accurately
repre-
sent
various
irregular
building
shapes
using
the
predefined
model
primitives.
Considering
the
complementary
advantages
of
LIDAR
data
and
high-resolution
imagery,
the
fusion
of
two
data
sources
is
regarded
as
a
promising
strategy
to
extract
high
quality
building
bound-
aries
[8,9,12–15,20–25].
However,
it
is
challenging
to
extract
the
correct
features
from
optical
image
or
LIDAR
data
of
complex
land-
scapes.
Additionally,
there
is
no
general
solution
for
fusing
different
features
for
automatic
building
boundary
extraction
of
various
building
shapes
such
as
curved,
wavy,
zigzag
and
other
irregular
shapes.
Most
existing
fusion
methods
can
only
handle
simple
build-
ing
shapes
like
polygons.
In
order
to
automatically
extract
building
boundaries
from
com-
plex
geographic
environments,
a
new
method
of
building
boundary
extraction
by
LIDAR-optical
fusion
is
proposed
in
this
paper.
This
0030-4026/$
–
see
front
matter ©
2013 Elsevier GmbH. All rights reserved.
http://dx.doi.org/10.1016/j.ijleo.2013.03.045