Multimedia Tools and Applications (2024) 83:43277–43302

https://doi.org/10.1007/s11042-023-17190-z

A systematic review on fall detection systems for elderly

healthcare

Archana Purwar1 · Indu Chawla1

Received: 25 December 2022 / Revised: 17 August 2023 / Accepted: 19 September 2023 /

Published online: 16 October 2023
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023

Abstract
To ensure healthy lives and promoting well-being for all in the society at all ages is one of
the goals of United Nations. Specially, health of elderly people plays an important factor in
productivity and prosperity of any country. According to reports, there will be over two bil-
lion elderly people worldwide by 2050. Most of elderly people live independently and need
some system to protect them from any kind of fall. As old people are highly susceptible
to fall due to weak body structure as well as some external conditions, researchers from
academia and industries are developing fall detections systems (FDS) or devices to prevent
them from fall. Hence, this paper majorly aims to review the papers on fall detection sys-
tems (FDS) to protect elderly people from any kind of fall. Papers selected for this study
spans from 2017- 2023. FDS will be helpful to sustain the health of elderly persons. In
view of strengthening research in this domain, this study gives an integrated and a critical
review of work done in this area for both wearable, non-wearable systems and hybrid sys-
tems with research directions as the advent of new technologies like deep learning, com-
puter vision, Internet of Things (IoT) and big data may improve the existing approaches/
systems.

Keywords Fall Detection systems · Elderly Healthcare · Machine learning · IoT ·

Segmentation · Wearable · Non-wearable · Sensors

1 Introduction

Each and every country all over the world is encountering an increase in population of
elderly people. As per World Health Organization (WHO), by 2030, one in six people
on the earth will be 60 or older [1]. Moreover, there will be twice as many people world-
wide who are 60 years of age or older in 2050. Ageing occurs biologically as a result

* Archana Purwar
archana.purwar@gmail.com
Indu Chawla
indu.chawla@jiit.ac.in
1
Department of Computer Science & Engineering and Information Technology, Jaypee Institute
of Information Technology, Noida, India

13
Vol.:(0123456789)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43278 Multimedia Tools and Applications (2024) 83:43277–43302

of the accumulation of several kinds of cellular and molecular damage over time. It
leads to weakening physical and mental abilities of older people. Consequently, elderly
people suffer from health related problems and keeping the people healthy is major con-
cern of all countries. Elderly people become more prone to falls due weak muscular
structure and other factors whose severity frequently necessitates medical treatment.
WHO estimates that 30% of persons over 65 experience one or more accidental falls
each year, and that incidence rises to 50% for those over 80. Due to this large number of
fall incidents, various methods described below have been developed to detect a number
of falls, prevent and protect senior people. [2, 3].

• Fall detection: Fall detection [4, 5] leads to methods which detect the happening of
fall. The systems operate on the principle of pattern recognition. In case of a sudden
change in the pattern, it works on human activity recognition like walking, sitting,
standing and notices sudden changes in the body sensor parameters and observes the
particular deviation as fall.
• Fall prevention: Fall prevention is one aspect in elderly healthcare [2]. The falls can
be prevented by avoiding risks thereby making home safer, going for regular health
checkups and right exercises, wearing comfortable clothing. It also involves generating
warning signals in case of possible fall to mitigate the falling risks.
• Fall protection: Fall protection [6–8] deals with arranging on time medical services.
Elderly can be protected from fall by doing for regular medical checkups. The home
environment should be safe like to avoid slippery floors of bathrooms, a bath mat
should be used to increase the grip, grab bars should be installed along with stairs, a
mobility aid like a simple cane stick can protect from possible fall. Moreover, in case
of fall, the medical emergency services should be informed on time and proper care
should be taken.

Abbreviations used in the paper are shown in Table 1. The following are the contri-
butions made in the manuscript:

• The manuscript has chosen to review articles related to Fall detection systems for
elderly people. Moreover, this paper has chosen a study of latest papers ranging from
2017–2023 in literature of fall detection systems to prevent elders from falls.
• A critical analysis of the recent articles is done as mentioned in the Table 2, 3, 4, 5 and
6.

Table 1  List of Abbreviations Abbreviations Description

and Acronyms
FDS Fall Detection Systems
IoT Internet of Things
SVM Support vector machine
LSTM Long Short-Term Memory networks
HAR Human Activity Recognition
FPS Fall Prevention Systems
FDAFPS Fall Detection and Prevention System
CNN Convolution Neural Network

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
 Table 2  Review of various methods for Camera-based fall detection
Author Year Dataset Purpose Methodology Result

Miguel et al. [42] 2017 53 videos were collected from a An accurate detection system Foreground Segmentation/ back- Sensitivity: 96%
laboratory and a house ground subtraction is used to Specificity: 96.9%
extract subject from images
Noise present in the data is
reduced using Kalman filter.
This filter is also employed to
trace periodic changes of differ-
ent human actions
KNN approach is applied to clas-
sify subject’s current state
Soni et al. [43] 2019 UR Fall Dataset ( publicly avail- Automatic fall detection for GMM based background subtrac- Sensitivity: 98.15%
able[77] indoor environment tion is done for detection of Specificity: 97.10%
movement of a person
Important geometric features are
extracted to distinguish fall from
Multimedia Tools and Applications (2024) 83:43277–43302

other actions of daily life using

Support vector machine (SVM)
Lezzar et al. [17] 2020 one public Fallen Person Dataset Improved Camera-based fall Fall, daily actions, and occlusion Sensitivity = 100%
(FPDS) [78], and the second detection method are distinguished by employing Precision = 93.94% for FPDS data
created locally SVM set
Features are extracted by a Yolov3
Kosarava et al 2021 Video Data set [77] and (ImViA) Simple fall control system Support vector machine, decision Accuracy: 100%
[79] [80]] tree and random forest were on both data sets
used as classification models
Keskes et al. [81] 2021 Data collected by using RGB-D Accurate and Robust fall detection Spatial Temporal Graph Convo- Accuracy:
camera system lutional Networks (ST-GCN) 100% on TST v2 Dataset
Publicly available data sets: TST using skeletal data was used 97.3% on Fallfree Dataset
Fall detection dataset v2, NTU 92.1% on NTU RGB-D Dataset
RGB-D,
and Fallfree dataset[82–84]
43279

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
 Table 2  (continued)
43280

Author Year Dataset Purpose Methodology Result

13
Shu & shu [85] 2021 Data set created Low cost Fall monitoring Calculation of velocities and Held-out accuracy
accelerations for fall is done : 89%
Relevance Vector Machine (RVM) Training accuracy: 94%
is taken for classification
Each color channel (red, green,
and blue) from each image is
used to obtain
Histogram of Oriented Gradients
features
Data matrix is made linear with
Principal Component Analysis
(PCA)
Zhao et al. [86] 2022 Publicly available data et: Accurate and light weight Fall Skeleton information is extracted Accuracy: 94.5% on NTU data set
NTU[82], UP- Fall[87] & UR- detection system using OpenPose and used to Accuracy: 98.85% on UP-Fall data
Fall [77] recognize fall set
Spatial information is used by Accuracy: 97% on UR data set
the adaptive graph convolution
module
A multi-scale temporal convolu-
tion network is employed to
fully get temporal contextual
details
Finally, recognition score is com-
puted using softmax function
Multimedia Tools and Applications (2024) 83:43277–43302

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 Table 2  (continued)
Author Year Dataset Purpose Methodology Result

Inturi et al. [88] 2023 Publicly available data set: UP- A fall and activity detection sys- Human joint points are extracted Accuracy: 98.59%
Fall[75] tem based on accurate keypoint using AlphaPose model Sensitivity: 94.37%
detection CNN is applied to analyze key- Specificity: 98.96%
points using spatial correlation
and These keypoints are inferred
to be the joint points of the
subject
Long short-term memory (LSTM)
is employed to protect long-term
dependencies
Jiangjiao et al. [89] 2022 Publicly available data set: UR[77] Improved fall detection system YOLOv3 for object identification Accuracy: 99.73% on UR data set
and Le2i [90] and Multi-stage Pose Estimation Accuracy: 99.62% on Le2i data set
Self created data set Network for finding ekypoints Accuracy: 94.74% on self created
are employed data set
Sliding window is used to divide
Multimedia Tools and Applications (2024) 83:43277–43302

to keypoints in frames to obtain

dynamic features
An adaptive keypoint attention
module is employed to improve
LSTM
Lian et al. [91] 2023 Publicly available data set: Robust fall detection system Weakly supervised learning-based Accuracy: 98.4% on UR data set
UR[77], Le2i [90], and dual-modal network is used Accuracy: 99.0% on Le2i data set
RFDS[91] A deep multiple instance learn- Accuracy: 97.2% on RFDS data set
ing framework is used to learn
the fall events using weak labels
43281

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
 Table 3   Pros-Cons of the various methods for Camera-based fall detection
43282

Reference Pros Cons

13
[42] • After fall is found, designed system shows images. To prevent any damage by third • It was designed for daylight situations; hence certain light and ambient conditions can
party, images can be modified. affect the quality of an image as well as gives more variance with time.
• Expensive camera were used to cover night time images may also worsen the quality of
an image.
• As elderly people have various kinds of movement patterns due to different health condi-
tions and mix of different age, nature of an observed subject must also be focused.
• Fall detection algorithms may get affected by any walking aids if they are in between
subject and camera.
[43] • Better detection rate as compared to state of art methods • Although static camera’s location was taken to minimize the occlusion, but in real world
occlusion may occur.
[17] • Improved camera-based fall detection systems • Study was fall detection in dark.
• Occlusions detection is done. • RGB cameras used could not detect objects in dark.
[79] • Computationally efficient without compromising accuracy • Occlusions were not taken care of.
[81] • Skeletal data is taken which has an edge over RGB data due to being invariant to illumi- • Partial and total occlusion scenarios were not taken care of.
nation and different backgrounds. • Chosen fall datasets is unrepresentative of real lif.e
• A trajectory of human motion using w ith a few joint positions are used by taking • Performance was not evaluated for an occlusion-robustness.
skeletal sequence.
• Low computational volume
[85] • Highly accurate, less expensive and multi-camera equipped fall detection system • Fall data set taken under study was limited as well as less diverse.
• Fall detection capable to cover long distant objects, capturing images in presence of • Shadow from humans was not considered in videos.
furniture/ people/animals in the same area
[86] • Light weight FDS • Skeleton information was not used for representation of falls.
• More Accurate • Approach did not play any role in fall events.
• Subgraph-based deep learning method
[88] • Accurate keypoints detection using AlphaPose as compared to OpenPose • Keypoint detection had not considered the issue of shadows which may result in wrong
• Computationally less complex due to use of a single camera-based system and RGB detection of keypoints.
images • Occlusion was also not taken care of in proposed method.
[89] • Improved performance in terms of accuracy • Takes more execution time
• Temporal features were not taken care of in proposed FDS.
[91] • Less execution time • Workload of data annotation is increased.
• No requirement of fine-grained annotations • Occlusion is also not taken care of in proposed method.
Multimedia Tools and Applications (2024) 83:43277–43302

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 Table 4  Contribution made by authors in the field of fall detection using wearable sensors
Author Dataset Purpose Methodology Result Pros Cons

Diana Yacchirema [55], SisFall public available Automatic activation IOT and Big data Accuracy = 91, 67% and Detects three types of Error rate of 8% in fall
2018 data set and notification of Precision = 93, 75% fall: forward, back- identification
alerts in case of fault ward and lateral
3D-axis accelerom- Smart IoT gateway for
eter embedded into a sending notifications
6LowPAN device for
data collection
Kai-Chun Liu [58], SisFall public data set Impact of sampling rate SVM, decision tree, For sampling rates Effects of sampling rate Different classification
2018 and one data created on energy savings naïve bayes, KNN of 11.6 and 5.8 Hz, from 200 to 128 are models performs best at
Minimum sampling rate Accuracy >  = 98% experimented using different sampling rate
that maintains high and 97% accuracy four machine learning
performance respectively methods
Luca Palmerini et al. 143 fall recordings from Multiphase fall model Support vector Sensitivity > 80%, a Good impact of feature More time in data acquisi-
[48], 2020 40 subjects from the features are more machines along with false alarm rate per extraction from a tion for multiphase
FARSEEING reposi- effective than conven- multiphase fall model hour = 0.56, and multiphase feature extraction
Multimedia Tools and Applications (2024) 83:43277–43302

tory tional features features F-measure = 64.6% fall model

Analysis of acceleration
signals received by
inertial sensor
Kai-Chun Liu [59], 494 falls from data Impact of window SVM, KNN, Regression Accuracy > 94% Comparison and effec- Difference in performance
2020 acquisition and Sisfall sizes on fall detection Tree, Naïve bayes tiveness of sliding is very less (within 1%)
dataset accuracy windows and impact
Segmentation defied windows as
approaches, sliding per the changes in
windows and impact- window sizes
defined windows, are
explored
Mustafa Sahin Turan 1600 fall trials Use of wearable motion Eight machine learning Accuracy > 96.2% Multiclass classification Time taken by best
et al. [92], 2021 sensors for finding fall classifiers indicating one out of performing algorithm is
direction four fall directions highest
Comparison of classifi-
cation algorithms
43283

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
 Table 4  (continued)
43284

Author Dataset Purpose Methodology Result Pros Cons

13
Marvi Waheed [61], SisFall and UP-Fall Data loss in Wearable Deep learning Accuracy >  = 97% Effectiveness in fall Comparison with Feature
2021 Detection Sensors detection in case selection techniques is
of noisy data using not done
Bidirectional Long
Short-Term Memory
(BiLSTM)
Qian et al. [93], 2022 Data collection done NB-IoT along with Mutilevel Threshold Accuracy = 94.88%, -portable and low-power Noise handling from sen-
MEMS Algorithms Sensitivity = 95.25%, fall detection system sor generated data is not
Microelectro- Specificity = 94.5% -multilevel threshold fall discussed
mechanical-systems detection algorithm
Le et al. [65], 2022 MobileAct 2.0 and UP- Time and frequency Classification algo- For MobileAct 2.0 Enhanced performance Performance is evaluated
Fall datasets based Feature extrac- rithms: SVM, k-NN, dataset: F1-Score due to effective feature on two datasets
tion along with Hjorth ANN, J48, and RF 95.23% (falls), 99.11% selection
parameters (non-falls),
For UP-Fall datasets:
96.16% (falls), 99.90%
(non-falls)
Butt [60], 2023 Fall data from Kaggle Machine learning LSTM and CNN-trans- Accuracy: LSTM = 88% Effectiveness of Deep Comparatively Longer
and UFRD dataset approaches fer learning CNN = 98% learning is discussed training time
Comparison of perfor- in fall detection
mance of LSTM and
CNN-transfer learning
Kulurkar [94], 2023 Real time data collec- low-cost wearable short-term with a long 95.87% accuracy Energy consumption Model is trained on small
tion sensing devices from memory network of wearable device is data size
Apache Flink and architecture taken into considera-
MbientLab tion
Multimedia Tools and Applications (2024) 83:43277–43302

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 Table 5  Summary of various sensors and devices suggested in different studies
Author, year Sensor Device Methodology

Amir mehmood, 2019 [95] Shimmer wearable sensors Waist band with shimmer sensor and adjustable Mahalanobis distance is used to measure
elastic strap similarity between two patterns to find fall
probability
Kimaya Desai, 2020 [40] Accelerometer, Gyroscope belt ML- Logistic Regression
Luca palmerini, 2020 [48] Inertial sensor attached on smartphone Smartphone fixed with belt worn on lower back SVM, KNN, Naïve bayes, random forests,
and Logistic regression
Al Nahin et al. [96] 3 d axis accelerometer embedded into a Accelerometer based Decision tree
6LowPAN device
Kai Chun Liu, 2018 [58] Accelerometer embedded in an Opal NA KNN, Naïve bayes, Decision Tree, SVM
Multimedia Tools and Applications (2024) 83:43277–43302

sensor
Mohammad Mehedi Hassan [56] Gyroscopes, accelerometers, and orien- Smartphones’ Hybrid framework for CNN-LSTM-Based
tation sensors deep learning
Samad Barri Khojasteh et al. [97] Accelerometer wrist-worn 3DACC on a smart wristband Genetic Algorithms (GA) and Simulated
Annealing (SA) for Threshold Optimization
Qian et al. [93] MEMS sensor Wristband having sensor Multilevel threshold algorithm
Honore [98] Smartphone in built pedometer virtual Smartphone Threshold-based accelerometer algorithm
sensor
Semwal [99] Inertial measurement unit (IMU) sensor smartphone 4 deep learning models, viz. CNN, LSTM,
present in smartphone gated recurrent unit (GRU), CNN + LSTM
Chunhua [100] thermopile Ambient sensor three-layer BP neural network
IR array sensor and MEMS PIR sensor
43285

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
 43286

Table 6  Analysis of Hybrid (Ccombination of wearable and non wearable) FDS

13
Reference Year Dataset Methodology Result Pros Cons

[56] 2019 MobiAct dataset Hybrid Deep CNN-LSTM- Accuracy >  = 96.75% mobile-enabled fall detec- Data set from only Acceler-
Based tion ometer is considered
[72] 2021 Self data collection Backpropagation neural Accuracy = 96.3%, specific- Reduction in response time Comparison with other
225 related to ADL and 180 network algorithm ity of 96.4%, Sensitivity approaches and datasets is
related to fall of 95.5% lacking
[70] 2019 Data captured through video cloud-network-edge archi- Accuracy = 96.67% Efficient in terms of accu- High computation time
and accelerometer tecture Recall = 96.67% Preci- racy
sion = 98.39% and
F1-score = 97.52%
[76] 2023 UP-Fall detection dataset[87] A multi-level feature fusion Accuracy: 97.9& when Cam- Efficiency of lHAR system A lot of resources are needed
technique is designed era 2 and wearable sensor is increased to train and infer results
A multi-head CNN with were used Not suitable for the use cases
attention module is Accuracy: 96.49% when such edge computing or
employed Camera 1 and wearable TinyML
Attention Module sensor were used
Convolutional LSTM is Accuracy: 94.9% when
used to deal with temporal camera 1 & camera 2 and
features wearable sensor were used
[101] 2020 Self created data set Employs a combination of Accuracy: 93.74% Less expensive solution Approach must be validated
YOLO-V3 and Lite- A mobile application on standard data set as well
FlowNet)
[102] 2022 UP-Fall dataset[87] Deep learning algorithms: Accuracy( CNN): 98.55% on Data set having inertial as Comparison with recent
DMLSmartActions dataset CNN, LSTM, ResNet and Up-Fall dataset well as video sensor approaches not given
Combined dataset others Accuracy(CNN): 86.97% on
DMLSmartActions
Accuracy (CNN): 92.86% on
combined data set
Multimedia Tools and Applications (2024) 83:43277–43302

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Multimedia Tools and Applications (2024) 83:43277–43302 43287

• Studies related to fall detection using wearable, non- wearable devices and hybrid
have been considered for this manuscript with future research directions.
• A systematic approach for doing review is done using PRISMA.

Remaining paper is organized into 7 sections. Second section illustrates related

work of review papers done by academicians for FDS and FPS. Section 3 gives the
procedure adopted for doing this review. Fourth, fifth and sixth section sections depict
an intensive review of research papers pertaining to non-wearable, wearable and hybrid
(fusion) FDS based on IoT, big data and cloud computing respectively. Section 7 gives
future research directions for fall detection systems. Lastly, Section 8 concludes the
paper.

2 Related work

Various researchers have done studies on fall detection and prevention systems
(FDAPS) [3, 9–12] using various technologies. Mooyeon et al. in [3] have discussed
how fall can be prevented using various applications. These applications deploy many
technologies such as video systems, virtual reality, artificial intelligence and Internet
of things (IoT) using wearable/non wearable devices, big data, virtual reality and oth-
ers. They discussed how fall can be reduced using these preventive methods. However,
they have not given detailed description of fall detection systems. Authors in [9] have
overviewed fall detection as well as fall prevention systems on various parameters such
as data sets, algorithms used, placement of sensors and age. However, their work lacks
those papers where deep learning algorithms were used. Marion, et al. in [10] discussed
various issues faced by researchers in designing FDAPS. Further, they also discussed
the difficulties such as digital divide, social stigma, setting threshold, entourage and
others faced by elderly people in adopting new technological applications to avoid falls.
But, a systematic review was lacking in their paper. Odasso, et al. of [11] have given
various ways and invention strategies to prevent fall. But they have not included the
study of fall detection systems. Another study done by Emily [12] revealed that fall
can be reduced by minimizing the risk of fall. They also exposed that how common
invention ways such as supportive footwear, eyeglass and education are not very much
effective for fall prevention. Recently, Torres-Guzman et al. [12] have gone through 44
papers and revealed that most of FDAPS are using smartphone and threshold- based
monitoring system. Their study did not cover the FDS comprising wearable andf non
wearable FDS. Alam et al. [13] has studied various papers related to only vision based
fall detection and Ramachandran eta l. [14] have given an overview of fall detection
systems using wearable devices.
Majority of the reviews done by researchers have got more papers of FDS as com-
pared to fall prevention systems (FPS) [3, 9–12, 15]. Therefore, this paper has chosen
to review literature on FDS, three types of solutions have been studied in the literature
of FDS. One is wearable devices, another is non wearable devices and third is hybrid
systems. Hence this paper has chosen to review articles related to FDS ranging from
2017–2023 in literature of fall detection systems to prevent elders from falls.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43288 Multimedia Tools and Applications (2024) 83:43277–43302

3 Proposed procedure of systematic review

A review presented under this study makes use of Preferred Reporting Items for Sys-
tematic Reviews and Meta-Analyses (PRISMA) [9, 16] technique during selection of
papers. The chosen approach for review comprises of identifying the papers for writing
review, selecting papers based on its suitability and finally including the leftover papers
for analysis. PRISMA technique is mainly divided into the following steps:

1. Identification
  In the beginning, around 35000 results from 2010 -2023 are obtained using the strings
“fall detection “ or “fall prevention” from academic library of Google scholar and IEEE-
explore. However, result set obtained was compressed by refining search by employing
multi-strings. 342 publications were set apart from databases. In first stage of PRISMA,
41 duplicate records obtained from both libraries are removed.
2. Screening
  During screening process, the some articles are also removed due to the following
reasons:
• 47 articles were removed due to Survey articles
• 29 articles were eliminated because of unavailability of articles from search engines
• 107 articles before 2016 were not considered.
• 92 articles not using machine learning/deep learning methods or not reported results
are also not reviewed.
3. Inclusion
  After screening process, critical analysis of 26 papers is included in this review. The
detailed analysis of these papers is given in Sections 4, 5 and 6.

4 Elderly healthcare using non‑wearable FDS

A great deal of work is being done by researchers to detect falls of elderly people using
non-wearable fall detection systems. The following sub sections summarize contribu-
tion made in directions of non-wearable devices, its effectiveness, and shortcomings of
these systems.

4.1 Contribution for non‑ wearable fall detection systems

These systems make use of cameras and various sensors like acoustic, environmental,
and infrared sensors. Review of these different FDS made by rsearchers is given below.

• Camera-based fall detection

Multiple cameras [4], a single camera, a 2D [17, 18] cameras,3D time of flight camera,
three-dimensional of images with depth data are all subcategories of camera-based systems
[19, 20]. The multi-camera system rebuilds a 3D image, evaluates the person’s volume dis-
tribution along the vertical axis, and notify when the majority of the volume is close to the
ground for a predetermined amount of time. This device is difficult to set up, takes lengthy

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43289

calibration, and is ineffective when there are multiple people present or when one is par-
tially blocked by furniture.
While time-of-flight cameras [19] are substantially more expensive and have lower lat-
eral resolution than conventional 2D video cameras. In contrast to wearable and ambient-
based detection systems, camera-based systems are still widely utilized because they pro-
vide various advantages in terms of robustness and the absence of human involvement after
installation. These devices are typically charged by power outlets or may be with a backup
power source (battery pack) [4].
Consequently, a thorough analysis of non-wearable systems is provided in the next para-
graph. Research community has devised a number of camera-based methods to identify the
fall. Table 2 and Table 3 gives detailed summary of work done by researchers along with
pros and cons of each method.

• Floor sensors/ambient sensors

There are context-aware systems that use a variety of sensors, including piezoelectric,
pressure [21], polymer, smart carpet, floor vibration sensors [22, 23] in addition to camera-
based systems. The ambient sensor network was set up to lower healthcare expenditures.
Health of human beings were monitored through periodic reporting, monitoring daily-life
actions, and various notifications. In another research work, throughout the house, various
sensors like magnetic contact, environmental, water, energy pressure, and passive infra-
red motion sensors were dispersed as part of the system [24].
When a subject walks on floor platforms or instrumented walkways, sensors were placed
along them to calculate gait using pressure/force sensors as well as moment transducers
[22, 25, 26]. Force platforms and pressure measurement systems are the two different cat-
egories of floor sensors [27, 28]. Although pressure sensors are able to find the centre of
pressure. These are incapable to compute the applied force vector. It should be separated
from force platforms. The pressure patterns under a foot can be measured using pressure
measurement instruments, but the horizontal or shear parts of the applied forces cannot be
utilized. The systems based on floor sensors analyzed the force exerted to the floor when
walking. Latest advances in this direction, propounds that future falls as well as adverse
incidents like physical functional decline [29–32] and fall risks [33, 34] in elderly health-
care can be forecasted by change in gait parameters.
Muheidat et al. [28] presented a context-aware and private real-time reporting aging in
place system. They designed a cooperative cloudlet system in which closest cloudlet will
receive the data from the sensors. Then, it will provide desired information in real time in
least possible time. Experiments obtain that their model is able to give 95% sensitivity and
85% specificity for detecting falls.
Recently, Alharthi [35] studied characteristics of gait changeability. They analyzed gait
intervals and found that these are accountable for various gait patterns in persons. Convo-
lution neural network (CNN) was employed in their floor sensor system.

• Acoustic sensors/infrared sensor

The sound of falls can be measured by employing acoustic sensors. An acoustic fall
detection system was designed by Khan et al. employing sound waves [36]. They extracted
Mel-scale frequency cepstral coefficients features and applied support vector machine.
Recently, Younis et al. proposed median deviated ternary patterns(features) to train SVM

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43290 Multimedia Tools and Applications (2024) 83:43277–43302

for classification of fall and non-fall incidents [37]. They evaluated performance of pro-
posed approach on A3 fall 2.0 dataset [38] and the MSP-UET fall detection datasets and
achieved an accuracy of 98% and 97%, respectively. Further, infrared sensors are also being
used for fall detection of humans [39].

4.2 Effectiveness of non‑wearable fall detection systems

Wearable devices like Tri-axial accelerometers need to be placed to the wrist or another
body part, or sewn into the fabric of shoes or clothing, to monitor body inclination [40].
The acceleration of rotation is calculated by gyroscopes [41]. However, the main issue
with this kind of technology is that elderly people frequently forget to wear them [17]
and in the instance of a help button, it is worthless if the individual has fallen asleep.
Additionally, these devices require batteries and an expensive 24-h monitoring staff in
addition to a monthly subscription charge [42].
These issues can be well handled with vision-based systems [4, 5, 17, 42–44]. The fun-
damental benefit of computer vision systems is that no extra equipment needs to be worn
by the user. From cameras, a lot of data may be gleaned, including position, motion, and
the subject’s movements. Therefore, a computer vision system not only provides informa-
tion on falls but also on other daily actions like the taking of medications or the timing of
meals and sleep [42].

4.3 Shortcomings of non‑wearable systems

Camera based systems need an installation cost [17, 42]. These devices should be placed as
well as positioned with care to take image / video of elderly person. This system also has
a privacy concerns. Moreover, in case of theft, these cameras can be switched off or bro-
ken; it will lead to unavailability of data. As a person can move from one place to another
place inside home, it necessitates multiple cameras to be installed with their capacity and
backup.
Sound sensors [36] in vision based systems also can malfunction due to less battery,
some perturbations and others. Infrared sensors [39] are influenced by hard articles like
smoke, dust, and others. It is also works work for short distances and not able to capture
data beyond that.

5 Elderly healthcare using wearable FDS

Falls suffered by elderly people may cause serious injuries. In that case, immediate
medical assistance is required. As a fall may occur at any place, so the wearable devices
or sensors are greatly beneficial for instant medical help. These devices present a cost
effective and easy to use system to identify fall based scenarios from other daily activi-
ties. The most effective fall detection system use machine learning algorithms. There
are different types of sensors that are used in wearable devices.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43291

• Smart sensors

Although, there are many sensors available and used, however all fall detection system
based on wearable devices use accelerometer, gyroscope in common [41].

• Accelerometer in wearable devices

Accelerometer is one of the most commonly used wearable sensor equipped in wide
range of fall detection systems. It captures body movement accelerations in three orthogo-
nal planes. These observation are related to a number of physical activities like step count,
running, time spent in various physical activities, energy expenditure etc. Wearable devices
like accelerometer have low accuracy sometimes while recording movements in adults
as compared to younger people as the movement through walker can be slow. This also
depends on the body part where sensor is placed. It can better capture if placed at hip
location as compared to hand part. To avoid this limitation, researchers have used research
grade accelerometer [Eduardo Teixeira]. Accelerometer offer several advantages like low
weight, less cost, low power consumption, small size, easy to use, can be embedded on any
device or can be mounted on different body parts.

• Gyroscopes

Another very popular sensor used for the purpose of fall detection is Gyroscope. Gyro-
scope is an inertial sensor which can measure the angular velocity and orientation of any
object and are also known angular sensors. The angular velocity is measured as the devia-
tion from the rotational angle of the object and depicted in degree per second. Gyroscope
and accelerometer are used together as the directional movement is measured by acceler-
ometer while any tilt or angular velocity is captured by gyroscope.

• Sensors embedded in smart phone

There are also studies that make use of sensors embedded in mobile phones [45–47].
Fall detection using smart devices are of added advantage. Luca palmerini et al. used iner-
tial sensor embedded on smartphone or a dedicated system. Both types of systems were
fixed and worn on the lower back. Smartphone was attached with a waist worn belt while
the other system was attached to the skin with the help of medical tape [48].
The shortcoming of these systems is that there is whole dependency on the mobile
phones. So, the concerned persons should always carry mobile phone with themselves and
do remember to keep it charged as well. Moreover, another difficulty is that the required
sensors are not always embedded with all kind of mobile phones. As a result, there may not
be effective results out of such system.
There is lot of contribution from authors who conducted experimentation for elderly fall
detection using wearable sensors. There is use of single sensor or multiple sensors in dif-
ferent proposed approaches. Study shows that utilizing signals from different sensors pro-
duce better and more accurate results. This survey presents different perspectives about the
elderly fall detection viz. data sources, variety of sensors and wearable devices. This study
will be helpful for the researchers who want to pursue work in elderly fall detection with
a summarization of recent work pursued in the field and with categorization of the some
of the points where further exploration can be advantageous. The work in the following

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43292 Multimedia Tools and Applications (2024) 83:43277–43302

subsections provides the literature review in order to investigate the present situation of
elderly fall detection using wearable devices.

5.1 Contribution in the field of fall detection with wearable sensors

In terms of supporting elderly healthcare, an accurate detection of fall incident is abso-

lutely essential to provide timely medical support. Numerous efforts have been put
and identified in the field of fall detection using wearable sensors [14, 49, 50]. Table 4
depicts the contribution made by authors in the field of fall detection using wearable
sensors along with pros and cons of each approach.

5.2 Fall detection using wearable sensors

Fall can be detected broadly in two ways using wearable sensors: one is by using threshold
based systems and second is by means of machine learning based approaches.

5.2.1 Threshold‑based wearable fall detection systems

Much work on fall detection has been proposed based on threshold based approach [103].
The application of threshold based approach has been proved useful in multiple aspects
like identification of fall scenario and classification of type of falls and near fall conditions
[51].
The threshold based method works on the principle that it detects a fall whenever the
value of acceleration obtained from the accelerometer which may be embedded in a wear-
able device is out of the boundary value of the threshold. Although this method seems very
simple, having less computational cost and complexity, however, the challenge is to figure
out the appropriate value for the threshold to distinguish daily activities from the fall.
According to Kimaya Desai et al., a sudden gradual decrease and then a subsequent
peak increase in the accelerometer value is considered as fall [40]. To get through sensor
integral errors, authors proposed an effective sensor fusion module which utilizes upper
and lower threshold values [52]. Fall detection system is implemented using differential
piezoresistive pressure sensors embedded in a carpet using threshold based technique [53]

5.2.2 Machine learning‑based wearable systems for fall detection

Although threshold based systems have been able to produce effective results in many stud-
ies. However, the approach could not produce correct result in some scenarios due to ambi-
guity in deciding the range of threshold. A number of machine learning approaches have
been proposed and applied to observe the corresponding effectiveness. De quadros et al.
proposed use of machine learning based approaches for identification of possible fall sce-
nario from the data obtained from wearable sensors embedded on wrist wearable device
[54]. A comparison between threshold based and machine learning approaches portray
that machine learning approaches produce better result as compared with threshold based
approaches. Figure 1 shows flow diagram for machine learning-based model building.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43293

Fig. 1  Flow diagram for machine learning-based model building

Data collection For a fall detection system using wearable devices, the features are
generally extracted from acceleration signals generated from accelerometer, pressure
sensor or gyroscope. Most commonly used features which are simple as well as useful for
fall detection are mean, standard deviation, tilt angles, sum vector magnitude etc. Static
activities like sitting and standing etc. can be detected by means of mean value. At the
same time, dynamic activities like running, walking, jumping etc. can be judged with the
help of standard deviation. Signal magnitude area is also used to distinguish between static
and dynamic activities. Other helpful features to identify static and dynamic activity are
calculation of angles between ground and device in addition to the angle between device
and gravitational vector. In the work done by Kimaya Desai et al., the data set consists
of readings from accelerometer and gyroscope along the three coordinate axes. For
identification of fall and no fall, readings from other daily activities like running, walking,
bending, jogging and squatting has been considered. Time window average technique is
used as the data from sensors belong to time series model. Diana Yacchirema et al. [55]
used two accelerometers and one gyroscope for data collection and another is publicly
available SisFall dataset for fall detection. The observations were gathered from 38
people. Out of these only observations for daily activity learning from 15 elderly people is
considered further.

Data pre‑processing Involves the processing and normalization of real time signals of
human activity as extracted with the help of sensors. Since the sensors data is a sequence
of samples, to analyze an activity, a windowing technique is applied. Different features like
acceleration, slop value is computed from features generated across three axes. Mohammad
Mehedi Hassan et al. calculated are 58 values for 20 statistical features computed for each
window frame [56]. Majd Saleh et al. applied two segment feature extraction method. Also
adopted an online method to consider the features with low computational cost [57].

Feature extraction For implementing fall detection system, the selection of distinctive
features from the sensor data obtained proves fruitful. There are many features which have
been considered by researchers. Authors performed fall detection using 54 features mainly
focussing on time domain statistical features employed to standard deviation, mean, skew-
ness of the three axes, and correlation coefficient between each pair of axes etc. [58, 59].
Another study was conducted on accelerometer data by extracting 44 features related to
Hjorth parameter, frequency domain and time domain [60].

Training The model is created and trained on a large set of labelled data. This trained
model is then used for testing the performance of proposed model. Different machine
learning algorithms have been proposed and used for training the dataset received from the
sensors and predicting the possibility of fall.

Wearable devices Authors have contributed in experimenting with use of wearable

devices. There are two purposes to use multiple devices and used at different parts of the

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43294 Multimedia Tools and Applications (2024) 83:43277–43302

body. The purpose is to figure out the user convenience as well as to get the appropriate
readings to access the correct prediction and assessment of fall.

Kimaya Desai et al. have suggested use of wearable belt for elderly people convenience.
It consists of a battery, micro-controller and MPU6050 and GSM module. The motion sen-
sors are placed at the front of the belt to capture the body orientation accurately.
Due to the necessity of timely intervention in case of elderly fall detection, there are fall
detection devices available in the market. This can be in the form of smartphone based fall
detection where inbuilt motion sensors try to distinguish the other physical activities from
fall. Table 5 depicts the various sensors and devices suggested in different studies.

5.2.3 Effectiveness and adoption of wearable sensors and devices

• Fall detection systems using wearable devices are more popular as compared to camera
based alternatives as these are low cost devices which does not interfere much with the
privacy of the user.
• The sensors are also able to monitor changes in the activities of daily living. From vio-
lent or agitated movements that can be identified as some signal of abnormal activities
happening to them like burglars etc.
• Other sensors like barometers, magnetometers, heart rate monitors, accelerometer and
gyroscope are generally found in most fall detection system based on sensors [41].
• The activities related to daily living are discriminated or distinguished by various types
of falls.(slips, trips, crashes, collapses etc.)
• There are no concerns of privacy issues while using devices embedded with sensor.

5.2.4 Shortcoming of wearable sensors and devices

Accelerometer and gyroscope are widely used in most of the research work for observation
purpose. But, it has a slow response time while gyroscopes have a fast response time.
The disadvantage of smart phone based system is that the user is supposed to carry
phone with him all day long. Second is the placement of smart phone based system.
According to experimental settings, if the position of smart phone changes from chest
pocket to pant pocket, it may not produce the same signals and the system may not perform
well.
It is difficult to test these systems in real time environment with elderly people. The
existing fall detection system mostly utilizes data collected from young adults as compared
to the elderly people as the real time data from elderly people is not available.
As the training and test data used in machine learning algorithms is chosen from the
same subject / person data. In actual scenarios, the subject will be different from the sample
data. Most of the time, the performance degrades when system is tested and trained on
different datasets. Some systems are not able to differentiate between the daily activities of
living and fall incident. Sometimes, wearable devices produce false alarms of emergency
and restrict user’s movements. The infrared sensors get impacted by temperature variation
and lighting conditions. Battery life of wearable devices is also limited and needs to be
recharged or changed which elderly people may forget to do.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43295

5.2.5 Methodologies adapted for fall detection for wearable devices

Machine learning algorithms Various machine learning algorithms viz Support vec-
tor machine (SVM), k-Nearest Neighbor (k-NN), Naïve Bayes (NB), Regression tree have
been widely applied in the current scenario.

Kai Chun Liu et al. applied in total, four machine learning algorithms to observe the per-
formance of the proposed system. Majd Saleh et al. applied machine learning algorithm,
Two SVM based fall detection algorithm is used to better achieve trade off among com-
plexity and accuracy. The first SVM is of low computational cost and high sensitivity
while the second one focuses more on accuracy.Activity of elders is captured through a 3
axial accelerometer [57]. Kimaya Desai et al. used Logistic regression predict the fall [40].
Diana Yacchirema et al. used decision tree model for fall detection [55]. Machine learning
models are very powerful and effective in detection of fall cases however the performance
lacks sometimes due to unbalanced and noisy nature of data obtained.
SVM is effective classification method but not suitable for handling large datasets due
to training complexity. KNN is simple to implement distance based algorithm however, it
faces issues while dealing with large dataset as it performs distance calculation from new
point to each already existing point. Naive bays algorithm can handle large datasets and
simple to implement. Variation in frequency distribution among training and test dataset
degrades the performance.

Low power wireless sensors network To overcome the limitations of non wearable sys-
tems, wearable systems have been proposed, which usually employs low power inertial
sensors like accelerometer and gyroscope typically attached to the body of the person for
movement recognition when a fall takes place.

Deep learning algorithms Deep learning approaches are very popular now a days due to
its capability to produce remarkable results. Deep learning methodology uses a large set
of labelled data and neural network architecture that contain many layers [104]. It allows
stacking of hidden layers to extract highly abstract features and make better re-use of
learned features. Marvi Waheed et al. used deep learning for fall detection using wearable
sensors [61]. Chen et al. also proposed use of deep learning on the data available by means
of wearable sensors for slow fall detection [62]. Jain et al. presented a pre fall detection
system to prevent the fall in order to mitigate the after effects of fall in elderly people. They
applied deep learning for identification of pre fall detection [63]. An improved fall detec-
tion model in terms of accuracy was proposed by using whale optimization along with
deep learning algorithm [64]. A comparison of deep learning algorithm using convolu-
tional neural network and long short term memory is performed in comparison to machine
learning algorithm on publicly available dataset [65]. Fall detection using range- Doppler
radar based has been demonstrated using deep learning approach [66].

Deep learning algorithms achieve higher accuracy as compared to machine learning

method or threshold based method. However, deep learning algorithms also require more
data for training.

Artificial neural network The use of artificial neural network has been studied by Casilari-
peraz and Francisco in possible fall detection. The study uses the data obtained from

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43296 Multimedia Tools and Applications (2024) 83:43277–43302

wearable devices for further investigation [67]. Luna-Perejon et al. also applied recurrent
neural network for fall detection in case of wearable sensor devices [68]. Fall detection using
CNN has been performed by using Wi-Fi-based CSI (Channel State Information) [69].

6 Elderly healthcare using hybrid FDS

These hybrid systems either using a combination of Internet of Things (IoT) [63], cloud
computing and big data technologies or a combination of wearable and non-wearable
systems.
Diana et al. applied smart IoT gateway, that enables the processing capabilities locally
with the intention of reducing the processing time. Even in case of fall, the emergency
alert notifications to healthcare professionals are sent through smart IoT gateway. It also
sends information related to type of fall along with the location of house of the elderly
person. Shahiduzzaman et al. projected the use of smart helmet for fall detection. Authors
proposed a novel cloud-network-edge architecture for the possible outcome [70]. Pal et al.
supported and presented the elderly healthcare by means of smart homes using Internet of
things [71]. Ng et al. also researched and experimented to identify incidences of fall using
IoT technology [72]. Deeppika et al. also proposed elderly healthcare application with the
help of wearable and non-wearable sources using IoT application [73]. To increase the effi-
ciency and accuracy of fall detection system, a hybrid approach has been proposed using
the deep learning algorithms on visual input captured by camera [74]. Deep learning algo-
rithms have been proposed on the images captured through CCTV camera to demonstrate
their applicability in real time scenarios [75].
A multi model approach has also been proposed recently which considers visual data
and sensor data to develop fusion architecture for human activity recognition. The visual
data is analyzed using Convolution Block Attention Module while multi source sensor data
is processed by using Convolutional Long Short Term Memory [76]. Summary of hybrid (a
combination of wearable and non wearable) FDS with prons and cons is given in Table 6.

Open problems Although, the existing work proposed various solution approaches for
effective fall detection and prevention. There are issues which need to be addressed in
future research. There is a strong need for designing low-cost wearable sensing devices
having less power consumption to increase battery life. The existing vision based
approaches lack in maintaining privacy and coverage area. Effective techniques and algo-
rithms need to be designed which takes care of real time management and support in case
of fall detection. Transfer learning can be applied to boost performance to overcome issues
of data unavailability.

7 Conclusion

Healthcare of human beings is one of the goals of all United Nations to provide happy,
peaceful and prosperous life for sustainable development. Moreover, elderly healthcare is
more challenging as older people suffers from a lot of health issues. Hence, this paper is

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43297

focusing on falls happening in elderly people so that appropriate action can be taken before
any mishap. In view of falls, a critical review of various recent studies is done for wearable
as well as non-wearable fall detection systems. Paper is concluded with future research
directions.

8 Future research directions for fall detection systems

Although, numerous approaches are proposed for detecting the fall for elderly health-
care using wearable as well as non-wearable devices. Still, there is a need to focus on
given issues.

• Although wearable device-based fall detection systems can recognize the human activ-
ity without compromising the user’s privacy but elderly people forgets to wear theses
devices. Hence, privacy secure vision-based system should be designed.
• Context aware systems represent the fall detection systems that uses sensors placed in
the areas such as pressure sensors, microphones and cameras. They have to be placed
at different places. Hence when users leave the area, it is impossible to capture data. It
leads to unavailability of data. Hence, robust fall detection systems should be designed.
• Fall may happen due to intrinsic (functional disability, balance impairments, vision.
Muscle, etc.) as well as extrinsic factors. It needs to develop systems focusing on reduc-
ing extrinsic risk factors which are of major concern.
• There are a few FDS in which experiments are made on a large and intensive real-life
dataset due to ethical reasons. Most FDS simulate fall like behavior in order to gather
various cases of fall events. Hence, we need more effective and reliable FPS in real life
settings.
• A few FDS have been designed to deal with occlusion. More FDS should be designed
to take care of the same.

Authors’ contributions Both the authors have contributed equally.

Funding No funding received.

Data availability Not Applicable.

Code availability Not Applicable.

Declarations
Ethics approval Not Applicable.

Consent to participate Not Applicable.

Consent for publication Not Applicable.

Conflicts of interest No conflict of interest.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43298 Multimedia Tools and Applications (2024) 83:43277–43302

References
1. https://​www.​who.​int/​news-​room/​fact-​sheets/​detail/​ageing-​and-​health
2. Hamm J, Money AG, Atwal A, Paraskevopoulos I (2016) Fall prevention intervention technolo-
gies: A conceptual framework and survey of the state of the art. J Biomed Inform 59:319–345
3. Oh-Park M et al (2021) Technology utilization in fall prevention. Am J Phys Med Rehabil
100(1):92–99
4. Rougier C et al (2011) Robust video surveillance for fall detection based on human shape defor-
mation. IEEE Trans Circ Syst Video Technol 21(5):611–622
5. Alanazi T, Babutain K, Muhammad G (2023) A robust and automated vision-based human
fall detection system using 3D multi-stream CNNs with an image fusion technique. Appl Sci
13(12):6916
6. Shi G et al (2009) Mobile human airbag system for fall protection using MEMS sensors and
embedded SVM classifier. IEEE Sensors J 9(9):495–503
7. Ding W, Chen X, Yu Z, Meng L, Ceccarelli M, Huang Q (2018) Fall protection of humanoids
inspired by human fall motion. In: 2018 IEEE-RAS 18th International Conference on Humanoid
Robots (Humanoids). IEEE, pp 827–833
8. Hu Z et al (2023) Impact behavior of nylon kernmantle ropes for high-altitude fall protection. J
Eng Fibers Fabr 18:155
9. Usmani S et al (2021) Latest research trends in fall detection and prevention using machine learn-
ing: A systematic review. Sensors 21(15):5134
10. Pech M et al (2021) Falls detection and prevention systems in home care for older adults: myth or
reality? JMIR Aging 4(4):e29744
11. Montero-Odasso M et al (2022) World guidelines for falls prevention and management for older
adults: a global initiative. Age Ageing 51(9):afac205
12. Harris E (2023) Systematic review: what works to prevent falls for older people. JAMA
13. Alam E et al (2022) Vision-based human fall detection systems using deep learning: A review.
Comput Biol Med 146:105626
14. Ramachandran A, Karuppiah A (2020) A survey on recent advances in wearable fall detection
systems. BioMed research international. 13:2020
15. Singh A et al (2020) Sensor technologies for fall detection systems: A review. IEEE Sensors J
20(13):6889–6919
16. Page MJ et al (2021) The PRISMA 2020 statement: an updated guideline for reporting systematic
reviews. Int J Surg 88:105
17. Lezzar F, Benmerzoug D, Kitouni I (2020) Camera-based fall detection system for the elderly with
occlusion recognition. Appl Med Inform 42(3):169–179
18. Hsu YW, Perng JW, Liu HL (2015) Development of a vision based pedestrian fall detection sys-
tem with back propagation neural network. In: 2015 IEEE/SICE international symposium on sys-
tem integration (SII). IEEE, pp 433–437
19. Diraco G, Leone A, Siciliano P (2010) An active vision system for fall detection and posture rec-
ognition in elderly healthcare. In: 2010 Design, Automation & Test in Europe Conference & Exhi-
bition (DATE 2010). IEEE, pp 1536–1541
20. Kepski M, Kwolek B (2014) Fall detection using ceiling-mounted 3d depth camera. In: 2014
International conference on computer vision theory and applications (VISAPP), vol 2. IEEE, pp
640–647
21. Youngkong P, Panpanyatep W (2021) A novel double pressure sensors-based monitoring and
alarming system for fall detection. In: In 2021 Second International Symposium on Instrumenta-
tion, Control, Artificial Intelligence, and Robotics (ICA-SYMP). IEEE, pp 1–5
22. Klack L, Möllering C, Ziefle M, Schmitz-Rode T (2011) Future care floor: A sensitive floor for
movement monitoring and fall detection in home environments. In: Wireless Mobile Commu-
nication and Healthcare: Second International ICST Conference, MobiHealth 2010, Ayia Napa,
Cyprus, October 18-20, 2010. Revised Selected Papers 1. Springer, Berlin Heidelberg, pp 211–218
23. Alwan M, Rajendran PJ, Kell S, Mack D, Dalal S, Wolfe M, Felder R (2006) A smart and passive
floor-vibration based fall detector for elderly. In: 2006 2nd International Conference on Informa-
tion & Communication Technologies, vol 1. IEEE, pp 1003–1007
24. VandeWeerd C, Yalcin A, Aden-Buie G, Wang Y, Roberts M, Mahser N, Fnu C, Fabiano D (2020)
HomeSense: Design of an ambient home health and wellness monitoring platform for older adults.
Health Technol 10(5):1291–1309. https://​doi.​org/​10.​1007/​s12553-​019-​00404-6
25. Kurita K (2012) Physical activity estimation method by using wireless portable sensor. In: SEN-
SORS, 2012 IEEE. IEEE, pp 1–4

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43299

26. Orr RJ, Abowd DG (2000) The smart floor: A mechanism for natural user identification and track-
ing. In: Extended abstracts on human factors in computing systems (CHI). ACM, New York, pp
275–276
27. Muro-De-La-Herran A, Garcia-Zapirain B, Mendez-Zorrilla A (2014) Gait analysis methods:
An overview of wearable and non-wearable systems, highlighting clinical applications. Sensors
14(2):3362–3394
28. Muheidat F, Lo’Ai AT (2020) In-home floor based sensor system-smart carpet-to facilitate healthy
aging in place (AIP). IEEE Access 8:178627–178638
29. Viccaro LJ, Perera S, Studenski SA (2011) Is timed up and go better than gait speed in predicting
health, function, and falls in older adults. J Am Geriatr Soc 59(5):887–892
30. Peel NM, Kuys SS, Klein K (2013) Gait speed as a measure in geriatric assessment in clinical set-
tings: A systematic review. J Gerontol Ser A 68(1):39–46
31. Van Kan GA, Rolland Y, Andrieu S, Bauer J, Beauchet O, Bonnefoy M, Cesari M, Donini LM, Gil-
lette-Guyonnet S, Inzitari M, Nourhashemi F, Onder G, Ritz P, Salva A, Visser M, Vellas B (2009)
Gait speed at usual pace as a predictor of adverse outcomes in community-dwelling older people an
international academy on nutrition and aging (IANA) task force. J Nutr Health Aging 13(10):881–889
32. Rochat S, Büla CJ, Martin E, Seematter-Bagnoud L, Karmaniola A, Aminian K, Piot-Ziegler C, San-
tos-Eggimann B (2010) What is the relationship between fear of falling and gait in well-functioning
older persons aged 65 to 70 years. Arch Phys Med Rehabil 91(6):879–884
33. Taylor ME, Ketels MM, Delbaere K, Lord SR, Mikolaizak AS, Close JCT (2012) Gait impairment
and falls in cognitively impaired older adults: An explanatory model of sensorimotor and neuropsy-
chological mediators. Age Ageing 41(5):665–669
34. Stone EE, Skubic M (2012) Capturing habitual, in-home gait parameter trends using an inexpensive
depth camera. In: 2012 Annual International Conference of the IEEE Engineering in Medicine and
Biology Society. IEEE, pp 5106–5109
35. Alharthi AS, Casson AJ, Ozanyan KB (2021) Spatiotemporal analysis by deep learning of gait signa-
tures from floor sensors. IEEE Sens J 21(15):16904–16914
36. Khan MS, Yu M, Feng P, Wang L, Chambers J (2015) An unsupervised acoustic fall detection system
using source separation for sound interference suppression. Signal Proc 110:199–210
37. Younis B, Javed A, Hassan F (2021) Fall detection system using novel median deviated ternary pat-
terns and SVM. In: 2021 4th International Symposium on Advanced Electrical and Communication
Technologies (ISAECT). IEEE, pp 01–05
38. Principi E, Droghini D, Squartini S, Olivetti P, Piazza F (2016) Acoustic cues from the floor: a new
approach for fall classification. Expert Syst Appl 60:51–61
39. Ben-Sadoun G, Michel E, Annweiler C, Sacco G (2022) Human fall detection using passive infrared
sensors with low resolution: a systematic review. Clin Interv Aging 17:35
40. Desai K, Mane P, Dsilva M, Zare A, Shingala P, Ambawade D (2020) A novel machine learning
based wearable belt for fall detection. In: 2020 IEEE International Conference on Computing, Power
and Communication Technologies (GUCON). IEEE, pp 502–505
41. Casilari E, Álvarez-Marco M, García-Lagos F (2020) A Study of the use of gyroscope measurements
in wearable fall detection systems. Symmetry 12(4):649
42. De Miguel K et al (2017) Home camera-based fall detection system for the elderly. Sensors
17(12):2864
43. Soni PK, Choudhary A (2019) Automated fall detection from a camera using support vector machine.
In: 2019 Second International Conference on Advanced Computational and Communication Para-
digms (ICACCP). IEEE, pp 1–6
44. Gutiérrez J, Rodríguez V, Martin S (2021) Comprehensive review of vision-based fall detection sys-
tems. Sensors 21(3):947
45. Biswas S, Bhattacharya T, Saha R (2018) On fall detection using smartphone sensors. In: 2018 Inter-
national conference on wireless communications, signal processing and networking (WiSPNET).
IEEE
46. Nguyen H, Zhou F, Mirza F, Naeem MA (2018) Fall detection using smartphones to enhance safety
and security of older adults at home. In: 2018 Eleventh International Conference on Mobile Comput-
ing and Ubiquitous Network (ICMU). IEEE, pp 1–2
47. Dogan JC, Hossain MS (2019) A novel two-step fall detection method using smartphone sensors. In:
2019 IEEE International Conference on Smart Computing (SMARTCOMP), IEEE, pp 434-438
48. Palmerini L, Klenk J, Becker C, Chiari L (2020) Accelerometer-based fall detection using machine
learning: Training and testing on real-world falls. Sensors 20(22):6479
49. Wang X, Ellul J, Azzopardi G (2020) Elderly fall detection systems: A literature survey. Front Robot
AI 7:71

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43300 Multimedia Tools and Applications (2024) 83:43277–43302

50. Hussain F et al (2019) Activity-aware fall detection and recognition based on wearable sensors. IEEE
Sensors J 19(12):4528–4536
51. Lee J-S, Tseng H-H (2019) Development of an enhanced threshold-based fall detection system using
smartphones with built-in accelerometers. IEEE Sens J 19(18):8293–8302
52. Sheikh SY, Jilani MT (2023) A ubiquitous wheelchair fall detection system using low-cost embedded
inertial sensors and unsupervised one-class SVM. J Ambient Intell Humaniz Comput 14(1):147–162
53. Chaccour K, Darazi R, el Hassans AH, Andres E (2015) Smart carpet using differential piezore-
sistive pressure sensors for elderly fall detection. In: 2015 IEEE ­11th International Conference on
Wireless and Mobile Computing, Networking and Communications (WiMob). IEEE, pp 225–229
54. de Quadros T, Lazzaretti AE, Schneider FK (2018) A movement decomposition and machine learn-
ing-based fall detection system using wrist wearable device. IEEE Sensors J 18(12):5082–5089
55. Yacchirema D, de Puga JS, Palau C, Esteve M (2018) Fall detection system for elderly people
using IoT and big data. Procedia Comput Sci 130:603–610
56. Hassan MM, Gumaei A, Aloi G, Fortino G, Zhou M (2019) A smartphone-enabled fall detection
framework for elderly people in connected home healthcare. IEEE Network 33(6):58–63
57. Saleh M, Jeannès RLB (2019) Elderly fall detection using wearable sensors: A low cost highly
accurate algorithm. IEEE Sensors J 19(8):3156–3164
58. Liu K-C, Hsieh C-Y, Hsu S-P, Chan C-T (2018) Impact of sampling rate on wearable-based fall
detection systems based on machine learning models. IEEE Sens J 18(23):9882–9890
59. Liu K-C, Hsieh C-Y, Huang H-Y, Hsu S-P, Chan C-T (2019) An analysis of segmentation
approaches and window sizes in wearable-based critical fall detection systems with machine learn-
ing models. IEEE Sens J 20(6):3303–3313
60. Butt A, Narejo S, Anjum MR, Yonus MU, Memon M, Samejo AA (2022) Fall detection using
LSTM and transfer learning. Wireless Pers Commun 126(2):1733–1750
61. Waheed M, Afzal H, Mehmood K (2021) NT-FDS—a noise tolerant fall detection system using
deep learning on wearable devices. Sensors 21(6):2006
62. Chen X, Jiang S, Lo B (2020) Subject-independent slow fall detection with wearable sensors via
deep learning. In: 2020 IEEE sensor. IEEE, pp 1–4
63. Valero CI et al (2021) AIoTES: Setting the principles for semantic interoperable and modern
IoT-enabled reference architecture for active and healthy ageing ecosystems. Comput Commun
177(2021):96–111
64. Eltahir MM, Yousif A, Alrowais F, Nour MK, Marzouk R, Dafaalla H, Hamza MA (2023) Deep
transfer learning-enabled activity identification and fall detection for disabled people. Comput
Mater Contin 75(2)
65. Le HL, Nguyen DN, Nguyen TH, Nguyen HN (2022) A novel feature set extraction based on
accelerometer sensor data for improving the fall detection system. Electronics 11(7):1030
66. Jokanović B, Amin M (2017) Fall detection using deep learning in range-Doppler radars. IEEE
Trans Aerosp Electron Syst 54(1):180–189
67. Casilari-Pérez E, García-Lagos F (2019) A comprehensive study on the use of artificial neural
networks in wearable fall detection systems. Expert Syst Appl 138:112811
68. Luna-Perejón F, Domínguez-Morales MJ, Civit-Balcells A (2019) Wearable fall detector using
recurrent neural networks. Sensors 19(22):4885
69. Nakamura T, Bouazizi M, Yamamoto K, Ohtsuki T (2020) Wi-fi-CSI-based fall detection by spec-
trogram analysis with CNN. In: GLOBECOM 2020-2020 IEEE Global Communications Confer-
ence, IEEE, pp 1–6
70. Shahiduzzaman KM, Hei X, Guo C, Cheng W (2019) Enhancing fall detection for elderly with
smart helmet in a cloud-network-edge architecture. In: 2019 IEEE International Conference on
Consumer Electronics-Taiwan (ICCE-TW). IEEE, pp 1–2
71. Pal D, Funilkul S, Charoenkitkarn N, Kanthamanon P (2018) Internet-of-things and smart homes
for elderly healthcare: An end user perspective. IEEE Access 6:10483–10496
72. Ng YJ, Anwar NSN, Ng WY, Law CQ (2021) Development of a fall detection system based on
neural network featuring IoT-technology. Int J Human Technol Interact (IJHaTI) 5(1):37–46
73. Deepika S, Vijayakumar KP (2022) IoT based elderly monitoring system. In: 2022 6th Interna-
tional Conference on Trends in Electronics and Informatics (ICOEI). IEEE, pp 573–579
74. Somkunwar RK, Thorat N, Pimple J, Dhumal R, Choudhari Y (2023) A novel based human fall
detection system using hybrid approach. J Data Acquis Process 38(2):3985
75. Sundaram BM, Rajalakshmi B, Mandal RK, Nair S, Choudhary SS (2023) Fall detection among
elderly using deep learning. In: 2023 International Conference on Intelligent and Innovative Tech-
nologies in Computing, Electrical and Electronics (IITCEE). IEEE, pp 554–558

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Multimedia Tools and Applications (2024) 83:43277–43302 43301

76. Islam MM, Nooruddin S, Karray F, Muhammad G (2023) Multi-level feature fusion for multi-
modal human activity recognition in Internet of Healthcare Things. Inf Fusion 94:17–31
77. Kwolek B, Kepski M (2014) Human fall detection on embedded platform using depth maps and
wireless accelerometer. Comput Methods Programs Biomed 117(3):489–501
78. Maldonado-Bascón S, Iglesias-Iglesias C, Martín-Martín P, Lafuente-Arroyo S (2019) Fallen peo-
ple detection capabilities using assistive robot. Electronics 8(9):915. https://​doi.​org/​10.​3390/​elect​
ronic​s8090​915
79. Kosarava K, Assanovich B (2021) A simple indoor fall control system for the elderly based on the
analysis of object bounding box parameters
80. Fall detection dataset, ImViA. https://​imvia.​ubour​gogne.​fr/​en/​datab​ase/​fall-​detec​tion-​datas​et-2.​
htm. Accessed 21 Jun 2021
81. Keskes O, Noumeir R (2021) Vision-based fall detection using ST-GCN. IEEE Access
9:28224–28236
82. Shahroudy A, Liu J, Ng TT, Wang G (2016) Ntu rgb+ d: A large scale dataset for 3d human activ-
ity analysis. In: Proceedings of the IEEE conference on computer vision and pattern recognition,
pp 1010–1019
83. Cippitelli E, Gambi E, Gasparrini S, Spinsante S (2016) TST fall detection dataset v2, IEEE Data-
port. IEEE. https://​doi.​org/​10.​21227/​H2VC7J
84. Alzahrani MS, Jarraya SK, Salamah MA, Ben-Abdallah H (2017) FallFree: Multiple fall scenario
dataset of cane users for monitoring applications using kinect. In: 2017 13th International Confer-
ence on Signal-Image Technology & Internet-Based Systems (SITIS). IEEE, pp 327–333
85. Shu F, Shu J (2021) An eight-camera fall detection system using human fall pattern recognition
via machine learning by a low-cost android box. Sci Rep 11(1):1–17
86. Zhao Z, Zhang L, Shang H (2022) A lightweight subgraph-based deep learning approach for fall
recognition. Sensors 22(15):5482
87. Martínez-Villaseñor L et al (2019) UP-fall detection dataset: A multimodal approach. Sensors
19(9):1988
88. Inturi AR, Manikandan VM, Garrapally V (2023) A novel vision-based fall detection scheme
using keypoints of human skeleton with long short-term memory network. Arab J Sci Eng
48(2):1143–1155
89. Li J et al (2022) KAMTFENet: a fall detection algorithm based on keypoint attention module and
temporal feature extraction. Int J Mach Learn Cybern 14(5):1831–1844
90. Charfi I, Miteran J, Dubois J, Atri M, Tourki R (201) Definition and performance evaluation of a
robust SVM based fall detection solution. In: 2012 eighth international conference on signal image
technology and internet based systems. IEEE, pp 218–224
91. Wu L et al (2023) Robust fall detection in video surveillance based on weakly supervised learning.
Neural Networks 163:286–297
92. Turan MŞ, Barshan B (2021) Classification of fall directions via wearable motion sensors. Digit
Signal Process 105:103129
93. Qian Z, Lin Y, Jing W, Ma Z, Liu H, Yin R, Zhang W (2022) Development of a real-time
wearable fall detection system in the context of Internet of Things. IEEE Internet Things J
9(21):21999–22007
94. Kulurkar P, Kumar Dixit C, Bharathi VC, Monikavishnuvarthini A, Dhakne A, Preethi P (2023)
AI based elderly fall prediction system using wearable sensors: A smart home-care technology
with IOT. Meas: Sensors 25:100614
95. Mehmood A, Nadeem A, Ashraf M, Alghamdi T, Siddiqui MS (2019) A novel fall detection algo-
rithm for elderly using SHIMMER wearable sensors. Heal Technol 9(4):631–646
96. Al Nahian MJ, Ghosh T, Al Banna MH, Aseeri MA, Uddin MN, Ahmed MR, Mahmud M, Kaiser
MS (2021) Towards an accelerometer-based elderly fall detection system using cross-disciplinary
time series features. IEEE Access 9:39413–39431
97. Khojasteh SB et al (2018) Improving fall detection using an on-wrist wearable accelerometer. Sen-
sors 18(5):1350
98. Honoré JT, Rask RD, Wagner SR (2022) Fall detection combining android accelerometer and step
counting virtual sensors. In: International Conference on ICT for Health, Accessibility and Well-
being. Springer Nature Switzerland, Cham, pp 3–16
99. Semwal VB, Kumar A, Nargesh P, Soni V (2023) Tracking of fall detection using IMU sensor: An
IoHT application. In: Machine Learning, Image Processing, Network Security and Data Sciences:
Select Proceedings of 3rd International Conference on MIND 2021. Springer Nature Singapore,
Singapore, pp 815–826

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
43302 Multimedia Tools and Applications (2024) 83:43277–43302

100. He C, Liu S, Zhong G, Wu H, Cheng L, Lin J, Huang Q (2023) A non-contact fall detection
method for bathroom application based on MEMS infrared sensors. Micromachines 14(1):130
101. Lv X, Gao Z, Yuan C, Li M, Chen C (2020) Hybrid real-time fall detection system based on deep
learning and multi-sensor fusion. In: 2020 6th International Conference on Big Data and Informa-
tion Analytics (BigDIA). IEEE, pp 386–391
102. Yergaliyev Z (2022) Human activity recognition and fall detection using video and inertial sensors
103. Torres-Guzman RA et al (2023) Smartphones and threshold-based monitoring methods effectively
detect falls remotely: A systematic review. Sensors 23(3):1323
104. Jain R, Semwal VB (2022) A novel feature extraction method for preimpact fall detection system
using deep learning and wearable sensors. IEEE Sens J 22(23):22943–22951

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

13
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center
GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers
and authorised users (“Users”), for small-scale personal, non-commercial use provided that all
copyright, trade and service marks and other proprietary notices are maintained. By accessing,
sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of
use (“Terms”). For these purposes, Springer Nature considers academic use (by researchers and
students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and
conditions, a relevant site licence or a personal subscription. These Terms will prevail over any
conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription (to
the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of
the Creative Commons license used will apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may
also use these personal data internally within ResearchGate and Springer Nature and as agreed share
it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not otherwise
disclose your personal data outside the ResearchGate or the Springer Nature group of companies
unless we have your permission as detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial
use, it is important to note that Users may not:

1. use such content for the purpose of providing other users with access on a regular or large scale
basis or as a means to circumvent access control;
2. use such content where to do so would be considered a criminal or statutory offence in any
jurisdiction, or gives rise to civil liability, or is otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association
unless explicitly agreed to by Springer Nature in writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a
systematic database of Springer Nature journal content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a
product or service that creates revenue, royalties, rent or income from our content or its inclusion as
part of a paid for service or for other commercial gain. Springer Nature journal content cannot be
used for inter-library loans and librarians may not upload Springer Nature journal content on a large
scale into their, or any other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not
obligated to publish any information or content on this website and may remove it or features or
functionality at our sole discretion, at any time with or without notice. Springer Nature may revoke
this licence to you at any time and remove access to any copies of the Springer Nature journal content
which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or
guarantees to Users, either express or implied with respect to the Springer nature journal content and
all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published
by Springer Nature that may be licensed from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a
regular basis or in any other manner not expressly permitted by these Terms, please contact Springer
Nature at

onlineservice@springernature.com