Currently,
the
3D
model
reconstruction
technology
based
on
binocular
stereo
vision
becomes
very
popular,
however,
current
matching
method
is
difficult
to
be
applied
for
objects
with
weakly-textured
surface,
further
leads
low
accuracy
and
poor
efficiency
of
those
objects.
To
improve
efficiency,
a
series
methods,
such
as
image
enhancement,
better
feature
extraction
algorithm
structured
light
have
been
searched.
However,
methods
provided
are
either
costly
or
computationally
complex,
which
lead
limited
application.
solve
above
problem,
new
has
proposed
in
this
paper,
which,
using
speckle
patterns
enhance
texture
features
surface.
In
addition,
analysis
traditional
takes
advantage
great
potential
deep
learning
techniques
efficiency.
Experiments
demonstrate
that
learning-based
network
achieves
10-3
pixels
9.37×105
POI/S.
With
method,
fast
accurate
surface
can
achieved.
Sensors,
Год журнала:
2025,
Номер
25(5), С. 1532 - 1532
Опубликована: Март 1, 2025
A
drawback
of
fringe
projection
profilometry
(FPP)
is
that
it
still
a
challenge
to
perform
efficient
and
accurate
high-resolution
absolute
phase
recovery
with
only
single
measurement.
This
paper
proposes
single-model
self-recovering
method
based
on
deep
learning.
The
built
Fringe
Prediction
Self-Recovering
network
converts
image
acquired
by
camera
into
four
mode
images.
algorithm
adopted
obtain
wrapped
phases
grades,
realizing
from
shot.
Low-cost
dataset
preparation
realized
the
constructed
virtual
measurement
system.
prediction
showed
good
robustness
generalization
ability
in
experiments
multiple
scenarios
using
different
lighting
conditions
both
physical
systems.
recovered
MAE
real
system
was
controlled
be
0.015
rad,
reconstructed
point
cloud
fitting
RMSE
0.02
mm.
It
experimentally
verified
proposed
can
achieve
under
complex
ambient
conditions.
Compared
existing
methods,
this
does
not
need
assistance
additional
modes
process
images
directly.
Combining
learning
technique
simplified
retrieval
unwrapping,
simpler
more
efficient,
which
provides
reference
for
fast,
lightweight,
online
detection
FPP.
Sensors,
Год журнала:
2025,
Номер
25(6), С. 1823 - 1823
Опубликована: Март 14, 2025
Fringe
projection
profilometry
(FPP)
is
a
measurement
technique
widely
used
in
the
field
of
3D
reconstruction.
However,
it
faces
issues
phase
shift
and
reduced
fringe
modulation
depth
when
measuring
translucent
objects,
leading
to
decreased
accuracy.
To
reduce
impact
surface
scattering
effects
on
wrapped
during
actual
measurement,
we
propose
single-frame
retrieval
method
named
GAN-PhaseNet
improve
subsequent
accuracy
for
objects.
The
network
primarily
relies
generative
adversarial
framework,
with
significant
enhancements
implemented
generator
network,
including
integrating
U-net++
architecture,
Resnet101
as
backbone
feature
extraction,
multilevel
attention
module
fully
utilizing
high-level
features
source
image.
results
ablation
comparison
experiment
show
that
proposed
has
superior
results,
not
only
achieving
conventional
objects
no
effect
slight
but
also
obtaining
lowest
errors
severe
compared
other
convolution
neural
networks
(CDLP,
Unet-Phase,
DCFPP).
Under
varying
noise
levels
frequencies,
demonstrates
excellent
robustness
generalization
capabilities.
It
can
be
applied
computational
imaging
techniques
field,
introducing
new
ideas
Optics Express,
Год журнала:
2024,
Номер
32(21), С. 36171 - 36171
Опубликована: Сен. 16, 2024
Simultaneous
structured
light
imaging
of
multiple
objects
has
become
more
demanding
and
widely
in
many
scenarios
involving
robot
operations
intelligent
manufacturing.
However,
it
is
challenged
by
pattern
aliasing
caused
mutual
reflection
between
high-reflective
objects.
To
this
end,
we
propose
to
learn
clear
fringe
patterns
from
aliased
mutual-reflective
observations
diffusion
models
for
achieving
high-fidelity
multi-body
reconstruction
line
with
typical
phase-shift
algorithms.
Regarding
as
a
formation
adding
significant
noise,
build
supervised
generative
learning
framework
based
on
then
train
self-attention-based
deep
network
U-Net-like
skip-connected
encoder-decoder
architecture.
We
demonstrate
the
generalization
capability
trained
model
recovery
its
performance
phase
three-dimensional
(3D)
shape
reconstruction.
Both
experimental
results
show
that
proposed
method
expected
feasibility
accuracy,
heralding
promising
solution
addressing
current
challenge
various
3D
tasks.
Photonics,
Год журнала:
2024,
Номер
11(11), С. 994 - 994
Опубликована: Окт. 22, 2024
Fringe
projection
profilometry
(FPP)
is
extensively
utilized
for
the
3D
measurement
of
various
specimens.
However,
traditional
FPP
typically
requires
at
least
three
phase-shifted
fringe
patterns
to
achieve
a
high-quality
phase
map.
In
this
study,
we
introduce
single-shot
method
based
on
common
path
polarization
interferometry.
our
method,
projected
pattern
created
through
interference
two
orthogonal
circularly
polarized
light
beams
modulated
by
liquid
crystal
spatial
modulator
(LC-SLM).
A
camera
employed
capture
reflected
pattern,
enabling
simultaneous
acquisition
four-step
phase-shifting
patterns.
The
system
benefits
from
advanced
anti-vibration
capabilities
attributable
self-interference
optical
design.
Furthermore,
utilization
low-coherence
LED
source
results
in
reduced
noise
levels
compared
laser
source.
experimental
demonstrate
that
proposed
can
yield
outcomes
with
high
accuracy
and
efficiency.