Applied Optics,
Год журнала:
2023,
Номер
62(9), С. 2178 - 2178
Опубликована: Фев. 13, 2023
The
measurement
model
of
binocular
vision
is
inaccurate
when
the
distance
much
different
from
calibration
distance,
which
affects
its
practicality.
To
tackle
this
challenge,
we
proposed
what
believe
to
be
a
novel
LiDAR-assisted
accuracy
improvement
strategy
for
visual
measurement.
First,
3D
points
cloud
and
2D
images
were
aligned
by
Perspective-n-Point
(PNP)
algorithm
realize
between
LiDAR
camera.
Then,
established
nonlinear
optimization
function
depth-optimization
lessen
error
depth.
Finally,
size
based
on
optimized
depth
built
verify
effectiveness
our
strategy.
experimental
results
show
that
can
improve
compared
three
stereo
matching
methods.
mean
decreased
33.46%
1.70%
at
distances.
This
paper
provides
an
effective
improving
Journal of Road Engineering,
Год журнала:
2024,
Номер
4(3), С. 257 - 281
Опубликована: Авг. 3, 2024
Automated
pavement
condition
survey
is
of
critical
importance
to
road
network
management.
There
are
three
primary
tasks
involved
in
surveys,
namely
data
collection,
processing
and
evaluation.
Artificial
intelligence
(AI)
has
achieved
many
breakthroughs
almost
every
aspect
modern
technology
over
the
past
decade,
undoubtedly
offers
a
more
robust
approach
automated
survey.
This
article
aims
provide
comprehensive
review
on
collection
systems,
algorithms
evaluation
methods
proposed
between
2010
2023
for
intelligent
In
particular,
system
includes
AI-driven
hardware
devices
vehicles.
The
including
right-of-way
(ROW)
cameras,
ground
penetrating
radar
(GPR)
devices,
light
detection
ranging
(LiDAR)
advanced
laser
imaging
etc.
These
different
components
can
be
selectively
mounted
vehicle
simultaneously
collect
multimedia
information
about
pavement.
addition,
this
pays
close
attention
application
artificial
detecting
distresses,
measuring
roughness,
identifying
rutting,
analyzing
skid
resistance
evaluating
structural
strength
pavements.
Based
upon
analysis
variety
state-of-the-art
methodologies,
remaining
challenges
future
needs
with
respect
discussed
eventually.
Journal of Construction Engineering and Management,
Год журнала:
2024,
Номер
150(7)
Опубликована: Май 3, 2024
Due
to
the
progress
in
light
detection
and
ranging
(LiDAR)
technology,
collection
of
road
point
cloud
data
containing
depth
information
spatial
coordinates
has
become
more
accessible.
Consequently,
utilizing
for
pavement
distress
quantification
emerges
as
a
crucial
approach
improving
precision
reliability
maintenance
procedures.
This
paper
aims
automatically
detect
visualize
using
LiDAR,
deep
learning-based
3D
object
method,
building
modeling
(BIM).
A
set
is
first
established
obtained
from
LiDAR.
Then,
network,
namely
PointPillar,
employed
detection,
results
will
be
quantified
at
region-level.
Finally,
BIM
model
integrating
parametrically
modeled
families
built
visually
manage
detected
distress.
After
training
validating
with
set,
performance
index
recall
78.5%,
mean
average
(mAP)
62.7%,
which
better
than
other
compared
cloud-based
methods
though
can
further
improved.
In
addition,
newly
untrained
section
applied
experiment.
The
integrated
environment
visual
management,
providing
guidance.
Applied Sciences,
Год журнала:
2024,
Номер
14(7), С. 2909 - 2909
Опубликована: Март 29, 2024
With
the
extension
of
road
service
life,
cracks
are
most
significant
type
pavement
distress.
To
monitor
conditions
and
avoid
excessive
damage,
crack
detection
is
absolutely
necessary
an
indispensable
part
periodic
maintenance
performance
assessment.
The
development
application
computer
vision
have
provided
modern
methods
for
detection,
which
low
in
cost,
less
labor-intensive,
continuous,
timely.
In
this
paper,
intelligent
model
based
on
a
target
algorithm
was
proposed
to
accurately
detect
classify
four
classes
cracks.
Firstly,
by
vehicle-mounted
camera
capture,
dataset
with
complicated
backgrounds
that
similar
actual
scenarios
built,
containing
4007
images
7882
samples.
Secondly,
YOLOv5
framework
improved
from
aspects
layer,
anchor
box,
neck
structure,
cross-layer
connection,
thereby
network’s
feature
extraction
capability
small-sized-target
were
enhanced.
Finally,
experimental
results
indicated
attained
AP
81.75%,
83.81%,
98.20%,
92.83%,
respectively,
mAP
89.15%.
addition,
achieved
2.20%
missed
rate,
representing
6.75%
decrease
over
original
YOLOv5.
These
demonstrated
effectiveness
practicality
our
addressing
issues
accuracy
small
targets
network.
Overall,
implementation
vision-based
models
can
promote
intellectualization
maintenance.
Structural Control and Health Monitoring,
Год журнала:
2024,
Номер
2024(1)
Опубликована: Янв. 1, 2024
Computer
vision
techniques
were
employed
to
monitor
the
displacement
of
retaining
walls
using
artificial
markers,
traditional
feature
detection
algorithms,
and
photogrammetry‐based
point
cloud
reconstruction.
However,
use
markers
often
increases
both
installation
time
costs,
whereas
performance
matching
is
affected
by
uneven
illumination,
photogrammetry
require
multiple
images
for
To
overcome
these
limitations,
a
nontarget‐based
monitoring
method
segmental
(SRWs)
combination
deep
learning
stereovision
was
proposed.
Binocular
reconstruct
geometry
surface
properties
SRW
in
digital
three‐dimensional
(3D)
model.
Deep
models
then
used
extract
natural
features
from
blocks,
enabling
calculation
without
targets.
The
evaluated
behaviors
experiments
at
laboratory
field
scales.
learning–based
image
segmentation
identified
block
experiment
real
case
datasets
with
an
average
F1
score
0.910
0.965
under
various
environmental
conditions.
reconstructed
results
coordinates
demonstrated
high
accuracy,
ranging
95.2%
98.6%.
Furthermore,
calculated
exhibited
degree
agreement
measured
displacement.
accuracy
displacements
ranged
89.5%
99.1%.
proposed
can
be
automatic
monitoring.
