The
goal
of
the
"Visual
Sentinel:
Video
Analytics
for
Missing
Subject
Identification"
project
is
to
automate
process
identifying
missing
subjects
from
CCTV
video
by
utilizing
cutting-edge
machine
learning
and
computer
vision
techniques.
One
project's
goals
create
a
real-time
surveillance
system
that
integrates
face
recognition
technology
increased
precision
accuracy.
has
ability
save
lives
bring
families
back
together
speeding
up
search
recovery
procedure.
To
effectively
locate
subjects,
uses
de-blurring
methods
algorithms.
matches
people
in
against
specified
data
collection,
giving
authorities
access
timely
information,
according
key
results.
sum
up,
Visual
Sentinel
provides
strong
practical
solution
enables
security
law
enforcement
professionals.
Scientific Reports,
Journal Year:
2023,
Volume and Issue:
13(1)
Published: Sept. 4, 2023
Abstract
Since
high-quality
real
data
acquired
from
selected
road
sections
are
not
always
available,
a
traffic
control
solution
can
use
software
simulators
working
offline.
The
results
show
that
in
contrast
to
microscopic
simulation,
the
algorithms
employing
neural
networks
work
real-time,
so
they
be
used,
among
others,
determine
speed
displayed
on
variable
message
signs.
This
paper
describes
an
experiment
develop
and
test
machine
learning
models,
i.e.,
long
short-term
memory,
gated
recurrent
unit
networks,
stacked
autoencoder
networks.
It
compares
their
effectiveness
with
prediction
generated
using
widely
recognized
simulator
analyzes
at
level
of
individual
vehicles.
IoT,
Journal Year:
2023,
Volume and Issue:
4(4), P. 486 - 513
Published: Oct. 24, 2023
The
falling
cost
of
IoT
cameras,
the
advancement
AI-based
computer
vision
algorithms,
and
powerful
hardware
accelerators
for
deep
learning
have
enabled
widespread
deployment
surveillance
cameras
with
ability
to
automatically
analyze
streaming
video
feeds
detect
events
interest.
While
analytics
is
currently
largely
performed
in
cloud,
edge
computing
has
emerged
as
a
pivotal
component
due
its
advantages
low
latency,
reduced
bandwidth,
enhanced
privacy.
However,
distinct
gap
persists
between
state-of-the-art
algorithms
successful
practical
implementation
edge-based
systems.
This
paper
presents
comprehensive
review
more
than
30
research
papers
published
over
last
6
years
on
(IE-SVA)
are
analyzed
across
17
dimensions.
Unlike
prior
reviews,
we
examine
each
system
holistically,
identifying
their
strengths
weaknesses
diverse
implementations.
Our
findings
suggest
that
certain
critical
topics
necessary
realization
IE-SVA
systems
not
sufficiently
addressed
current
research.
Based
these
observations,
propose
trajectories
short-,
medium-,
long-term
horizons.
Additionally,
explore
trending
other
areas
can
significantly
impact
evolution
Developing
and
designing
border
surveillance
systems
to
meet
specific
needs
requirements
is
a
comprehensive
process
that
involves
careful
assessment,
customization,
consideration
of
environmental
operational
factors.
These
are
crucial
for
most
nations
worldwide,
as
they
can
provide
real-time
monitoring
vast
areas,
including
remote
challenging
terrains,
which
might
pose
challenge
systems.
Covering
extensive
areas
with
difficult
terrains
presents
significant
challenges
due
power
limitations
high
costs.
To
address
the
faced
by
current
systems,
such
limitations,
alternative
energy
sources
energy-efficient
technologies
proposed.
Additionally,
cost
management
achieved
through
selection
equipment
modular
designs.
This
enables
terrestrial
environments
operate
effectively
in
large,
remote,
terrains.
The
thesis's
contribution
field
focuses
specifically
on
Libyan
Desert
border.
It
highlights
unique
approach
research
its
relevance
addressing
geographical
context.
system
utilizes
unmanned
fixed
platforms
equipped
infrared
cameras
(FLIR)
employs
edge
computing
Internet
Things
(IoT)
framework.
Within
this
framework,
two
Automatic
Target
Recognition
(ATR)
based
machine
learning
algorithms,
Bag-of-Features
feature
extraction
supervised
classification,
implemented.
run
low-power
microprocessors
capacity
IoT
nodes.
first
proposed
segments
image
into
regions
interest
before
processing,
while
second
ATR
works
directly
entire
image.
evaluate
performance
these
dataset
images
relevant
Sahara
environment
used,
allowing
testing
assessment
system's
capabilities.
In
evaluation
process,
both
approaches
considered
combination
four
different
classification
algorithms:
Support
Vector
Machine
(SVM),
K-Nearest
Neighbors
(KNN),
Decision
Tree
(DT),
Naive
Bayes
(NB),
along
three
descriptors:
SURF,
SIFT,
ORB.
As
conclusion
experimental
use
generic
classes
recommended
low
resolution
images.
SURF-SVM
prediction
highest
detection
capacity,
reaching
up
97\%,
frame
rates
5.71
device
59.17
workstation.
approach,
focusing
classifying
categories
(animal,
vehicle,
person),
resulted
reduced
confusion
between
compared
identifying
targets.
By
employing
categories,
an
increase
capacity.
results
demonstrate
feasibility
using
surveillance,
even
like
Desert.
RESUMEN
Desarrollar
y
diseñar
sistemas
de
vigilancia
fronteriza
para
cumplir
con
necesidades
requisitos
específicos
es
un
proceso
integral
que
implica
una
evaluación
cuidadosa,
personalización
consideración
factores
ambientales
operativos.
Estos
son
importantes
la
mayoría
las
naciones
en
todo
el
mundo,
ya
pueden
proporcionar
monitoreo
tiempo
real
grandes
áreas,
incluyendo
terrenos
remotos
difíciles,
podrían
ser
desafío
los
vigilancia.
Cubrir
áreas
difíciles
plantea
desafíos
significativos
debido
limitaciones
energía
altos
costos.
Para
abordar
enfrentados
por
actuales
fronteriza,
como
energía,
se
proponen
fuentes
alternativas
tecnologías
eficientes
además
gestionar
costos
mediante
selección
cuidadosa
equipos
diseños
modulares.
Esto
permite
entornos
terrestres
operen
manera
efectiva
grandes,
difíciles.
La
contribución
tesis
al
campo
centra
específicamente
frontera
del
Desierto
Libio.
Destaca
enfoque
único
investigación
su
relevancia
contexto
geográfico
ambiental
específico.
El
sistema
utiliza
plataformas
fijas
no
tripuladas
equipadas
cámaras
infrarrojos
emplea
computación
borde
Cosas.
Dentro
este
marco,
implementan
dos
Reconocimiento
Automático
Objetivos
basados
algoritmos
aprendizaje
automático,
extracción
características
uso
clasificación
supervisada.
ejecutan
microprocesadores
baja
potencia
capacidad
cómputo
nodos
primer
propuesto
segmenta
imagen
regiones
interés
antes
procesamiento,
mientras
segundo
trabaja
directamente
completa.
evaluar
rendimiento
estos
conjunto
datos
imágenes
relevante
entorno
Sáhara,
permitiendo
pruebas
exhaustivas
capacidades
sistemas.
llevar
cabo
evaluación,
consideraron
ambas
aproximaciones
combinación
cuatro
diferentes,
Máquina
Soporte
Vectorial
Vecinos
Más
Cercanos
Árbol
Decisiones
(DT)
tres
descriptores,
SIFT
Como
conclusión
experimental,
recomienda
utilizar
clases
genéricas
resolución
infrarrojos.
Además,
predicción
basado
logró
mayor
detección,
alcanzando
hasta
dispositivo
Cosas
estación
trabajo.
Este
enfoque,
clasificar
categorías
vehículo
persona),
resultó
disminución
confusión
entre
comparación
identificación
objetivos
específicos.
Al
emplear
genéricas,
aumento
detección.
resultados
muestran
viabilidad
incluso
desafiantes
Sáhara.
Research Square (Research Square),
Journal Year:
2024,
Volume and Issue:
unknown
Published: Oct. 25, 2024
Abstract
Addressing
the
challenges
of
time-consuming
and
labor-intensive
traffic
data
collection
annotation,
along
with
limitations
current
deep
learning
models
in
practical
applications,
this
paper
proposes
a
cross-domain
object
detection
transfer
method
based
on
digital
twins.
A
twin
scenario
is
constructed
using
simulation
platform,
generating
virtual
dataset.
To
address
distributional
discrepancies
between
real
datasets,
multi-task
algorithm
graph
matching
introduced.
The
employs
module
to
align
feature
distributions
source
target
domains,
followed
by
network
for
detection.
An
attention
mechanism
then
applied
instance
segmentation,
two
tasks
exhibiting
different
noise
patterns
that
mutually
enhance
robustness
learned
representations.
Additionally,
multi-level
discriminator
designed,
leveraging
both
low-
high-level
features
adversarial
training,
thus
enabling
share
useful
information,
which
improves
performance
proposed
tasks.
