Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies,
Год журнала:
2025,
Номер
9(1), С. 1 - 38
Опубликована: Март 3, 2025
Human
activity
recognition
(HAR)
using
ambient
sensors
in
smart
homes
has
numerous
applications
for
human
healthcare
and
wellness.
However,
building
general-purpose
HAR
models
that
can
be
deployed
to
new
home
environments
requires
a
significant
amount
of
annotated
sensor
data
training
overhead.
Most
vary
significantly
their
layouts,
i.e.,
floor
plans
the
specifics
embedded,
resulting
low
generalizability
trained
specific
homes.
We
address
this
limitation
by
introducing
novel,
layout-agnostic
modeling
approach
systems
utilizes
transferrable
representational
capacity
natural
language
descriptions
raw
data.
To
end,
we
generate
Textual
Descriptions
Of
Sensor
Triggers
(TDOST)
encapsulate
surrounding
trigger
conditions
provide
cues
underlying
activities
models.
Leveraging
textual
embeddings,
rather
than
data,
create
predict
standard
across
without
(re-)training
or
adaptation
target
Through
an
extensive
evaluation,
demonstrate
effectiveness
TDOST-based
unseen
through
experiments
on
benchmark
Orange4Home
CASAS
datasets.
Furthermore,
conduct
detailed
analysis
how
individual
components
our
affect
downstream
performance.
Sensors,
Год журнала:
2023,
Номер
23(13), С. 6040 - 6040
Опубликована: Июнь 29, 2023
Smart
living,
an
increasingly
prominent
concept,
entails
incorporating
sophisticated
technologies
in
homes
and
urban
environments
to
elevate
the
quality
of
life
for
citizens.
A
critical
success
factor
smart
living
services
applications,
from
energy
management
healthcare
transportation,
is
efficacy
human
action
recognition
(HAR).
HAR,
rooted
computer
vision,
seeks
identify
actions
activities
using
visual
data
various
sensor
modalities.
This
paper
extensively
reviews
literature
on
HAR
amalgamating
key
contributions
challenges
while
providing
insights
into
future
research
directions.
The
review
delves
essential
aspects
state
art
potential
societal
implications
this
technology.
Moreover,
meticulously
examines
primary
application
sectors
that
stand
gain
such
as
homes,
healthcare,
cities.
By
underscoring
significance
four
dimensions
context
awareness,
availability,
personalization,
privacy
offers
a
comprehensive
resource
researchers
practitioners
striving
advance
applications.
methodology
involved
conducting
targeted
Scopus
queries
ensure
coverage
relevant
publications
field.
Efforts
have
been
made
thoroughly
evaluate
existing
literature,
gaps,
propose
comparative
advantages
lie
its
addressing
limitations
previous
offering
valuable
Smart
living,
a
concept
that
has
gained
increasing
attention
in
recent
years,
revolves
around
integrating
advanced
technologies
homes
and
cities
to
enhance
the
quality
of
life
for
citizens.
Sensing
human
action
recognition
are
crucial
aspects
this
concept.
living
applications
span
various
domains,
such
as
energy
consumption,
healthcare,
transportation,
education,
which
greatly
benefit
from
effective
recognition.
This
field,
originating
computer
vision,
seeks
recognize
actions
activities
using
not
only
visual
data
but
also
many
other
sensor
modalities.
paper
comprehensively
reviews
literature
on
smart
environments,
synthesizing
main
contributions,
challenges,
future
research
directions.
review
selects
five
key
domains:
Technology,
Multimodality,
Real-time
Processing,
Interoperability,
Resource-Constrained
they
encompass
critical
required
successfully
deploying
living.
These
domains
highlight
essential
role
sensing
play
developing
implementing
solutions.
serves
valuable
resource
researchers
practitioners
seeking
explore
further
advance
field
Information Fusion,
Год журнала:
2023,
Номер
104, С. 102197 - 102197
Опубликована: Дек. 16, 2023
Wearable
devices
and
smart
sensors
are
increasingly
adopted
to
monitor
the
behaviors
of
human
artificial
agents.
Many
applications
rely
on
capability
such
recognize
daily
life
activities
performed
by
monitored
users
in
order
tailor
their
with
respect
occurring
situations.
Despite
constant
evolution
sensing
technologies
numerous
research
this
field,
an
accurate
recognition
in-the-wild
situations
still
represents
open
challenge.
This
work
proposes
a
novel
approach
for
situation
identification
capable
recognizing
which
they
occur
different
environments
behavioral
contexts,
processing
data
acquired
wearable
environmental
sensors.
An
architecture
situation-aware
computing
system
is
proposed,
inspired
Endsley's
situation-awareness
model,
consisting
two-step
identification.
The
first
identifies
via
learning-based
technique.
Simultaneously,
context
recognized
using
Context
Space
Theory.
Finally,
fusion
between
state
allows
identifying
complex
user
acting.
knowledge
regarding
forms
basis
smarter
can
be
realized.
has
been
evaluated
ExtraSensory
public
dataset
compared
state-of-the-art
techniques,
achieving
accuracy
96%
significantly
low
computational
time,
demonstrating
efficacy
approach.
Deep
learning-based
Human
Activity
Recognition
(HAR)
systems
received
a
lot
of
interest
for
health
monitoring
and
activity
tracking
on
wearable
devices.
The
availability
large
representative
datasets
is
often
requirement
training
accurate
deep
learning
models.
To
keep
private
data
users'
devices
while
utilizing
them
to
train
models
huge
datasets,
Federated
Learning
(FL)
was
introduced
as
an
inherently
distributed
paradigm.
However,
standard
FL
(FedAvg)
lacks
the
capability
heterogeneous
model
architectures.
In
this
paper,
we
propose
via
Augmented
Knowledge
Distillation
(FedAKD)
FedAKD
evaluated
two
HAR
datasets:
A
waist-mounted
tabular
dataset
wrist-mounted
time-series
dataset.
more
flexible
than
federated
it
enables
collaborative
with
various
capacities.
considered
experiments,
communication
overhead
under
200X
less
compared
methods
that
communicate
models'
gradients/weights.
Relative
other
model-agnostic
methods,
results
show
boosts
performance
gains
clients
by
up
20
percent.
Furthermore,
shown
be
relatively
robust
statistical
scenarios.
Smart
Living,
an
increasingly
prominent
concept,
entails
incorporating
sophisticated
technologies
in
homes
and
urban
environments
to
elevate
the
quality
of
life
for
citizens.
A
critical
success
factor
Living
services
applications,
from
energy
management
healthcare
transportation,
is
efficacy
human
action
recognition
(HAR).
HAR,
rooted
computer
vision,
seeks
identify
actions
activities
using
visual
data
various
sensor
modalities.
This
paper
extensively
reviews
literature
on
HAR
amalgamating
key
contributions
challenges
while
providing
insights
into
future
research
directions.
The
review
delves
essential
aspects
state
art
potential
societal
implications
this
technology.
Moreover,
meticulously
examines
primary
application
sectors
that
stand
gain
such
as
smart
homes,
healthcare,
cities.
By
underscoring
significance
four
dimensions
Context
Awareness,
Data
Availability,
Personalization,
Privacy
serves
a
valuable
resource
researchers
practitioners
striving
advance
applications.
Because
of
the
speedy
surge
in
social
media's
expansion,
proliferation
malicious
and
harmful
poses
a
substantial
worry
contemporary
society.
The
identification
hate
speech
on
platforms
like
Twitter
is
crucial
for
various
tasks
such
as
controversial
event
extraction,
AI
chatterbot
creation,
content
suggestions,
sentiment
analysis.
Researchers
have
invested
considerable
effort
addressing
challenging
task
identifying
hostile
due
to
rise
information.
objective
classify
tweets
Hateful,
Offensive,
or
neither.
However,
this
highly
complex
intricate
nature
natural
language
constructs,
encompassing
different
manifestations
animosity
directed
at
demographics,
multitude
ways
same
meaning
can
be
expressed.Previous
research
has
predominantly
relied
manual
feature
extraction
employed
representation-learning
techniques
followed
by
linear
classifiers.
Nevertheless,
deep
learning
methods
recently
demonstrated
significant
accuracy
improvements
problems
across
speech,
vision,
text
applications.
In
study,
This
paper
present
an
idea
automatic
classifications
inappropriate
expressions
hostility
using
transfer
models.
research,
leverage
classified
tweet
datasets
obtained
from
Kaggle
conduct
experiments.
Findings
reveal
that
multilingual-BERT
model
its
pre-trained
version
deliver
superior
outcomes.
Specifically,
BERT
notably
improves
classification
hateful
up
92%
when
compared
other
algorithms.
Algorithms,
Год журнала:
2024,
Номер
17(10), С. 434 - 434
Опубликована: Сен. 28, 2024
Human
Activity
Recognition
(HAR)
is
a
rapidly
evolving
field
with
the
potential
to
revolutionise
how
we
monitor
and
understand
human
behaviour.
This
survey
paper
provides
comprehensive
overview
of
state-of-the-art
in
HAR,
specifically
focusing
on
recent
techniques
such
as
multimodal
techniques,
Deep
Reinforcement
Learning
large
language
models.
It
explores
diverse
range
activities
sensor
technologies
employed
for
data
collection.
then
reviews
novel
algorithms
used
emphasis
multimodality,
gives
an
datasets
physiological
data.
also
delves
into
applications
HAR
healthcare.
Additionally,
discusses
challenges
future
directions
this
exciting
field,
highlighting
need
continued
research
development
fully
realise
various
real-world
applications.