1

Challenges, Opportunities and Trends in Smart Fall

Detection Systems: A User Centric Approach
Tuan Nguyen Gia1 , Minh Nguyen4 , Imed Ben Dhaou1,2,3 ,Tomi Westerlund1
1
Department of Future Technologies, University of Turku, Turku, Finland
2
Department of Electrical Engineering, College of Engineering, Qassim University, Unaizah
3
Univesity of Monastir, Tunisia
Email: [email protected], [email protected], [email protected]

Abstract—Falls can cause serious traumas such as brain these systems, Internet-of-Things (IoT), that can be described
injuries and bone fractures, especially among elderly people. Fear as interconnected networks where humans and objects from
of falling might reduce physical activities resulting in declining different disciplines can interact and communicate with each
social interactions and eventually causing depression. To lessen
the effects of a fall, timely delivery of medical treatment can other, is one of the main technologies [10], [11]. An IoT fall
play a vital role. In a similar scenario, an IoT-based wearable detection system can collect bio-signals and motion-related
system can provide the most promising way to mitigate serious data such as (e.g., heart rate, respiration rate, acceleration
consequences of a fall while providing the convenience of usage. and angular velocity of the body) and contextual data (e.g.,
However, to deliver sufficient degree of monitoring and reliability, falling situation, location and surrounding environment). The
wearable devices working at the core of a fall detection systems
are required to work for a prolonged period of time. Additionally, collected data can be analyzed and processed in real-time
the selection of the most suitable wearable technologies should for extracting useful information. As a consequence, medical
account for the perception of the end user such as doctor, caregivers can be informed in real-time and they can monitor
healthcare giver and patient. This paper compared and contrasts the information related to the fall cases that help to improve
existing IoT-enabled fall-detection systems, presents the user quality of treatment.
perception of the fall detection system, and .
Although the recent works have discussed about these
Index Terms—Internet-of-Things, IoT, Fall Detection, Energy systems, they were not deeply investigated in terms of chal-
Efficiency, Wearable Devices, accelerometer, gyroscope, magne- lenges, opportunities, trends and user opinions. In [12], fall
tometer, nRF
detection systems were categorized into three groups based
on wearable devices, ambiance sensors, and cameras. Dif-
I. I NTRODUCTION ferently, we believe that IoT fall detection systems can be
Fall cannot be underestimated as fall is the second leading categorized into five types including wearable-based, camera-
cause of injury deaths worldwide. It is estimated that more based, wireless-based, contextual sensor based, and hybrid-
than 646 thousand people die from falls globally every year based systems in which the hybrid-based systems utilize
[1]. Fall can occur because of subjective or objective reasons. several technologies (e.g., wearable and contextual sensors)
For example, a person can fall due to a stroke or a car accident. simultaneously to improve the accuracy of fall detection. Each
As a result, it can cause serious injuries such as traumatic brain group has its own advantages and disadvantages. For example,
damages, head traumas, and hip fracture [2], [3]. More than wireless-based IoT fall detection systems often respect data
37 million severe falls requiring medical care occur every year privacy as sensitive information is not shown while some
[1]. Treating these serious injuries from falls can last around other systems such as camera-based systems often do not
several months and is costly (e.g., approximately 30 thousand consider data privacy properly. Depending on the case, a
US dollars for a serious case) [4], [5]. Anyone in any ages specific fall detection system is more preferable. In this paper,
with any health status can fall. However, children and elderly we present the general IoT fall detection system architecture
people over 65 years of age are likely to fall easier. Similarly, and discuss different types of IoT fall detection systems.
people with bone-related diseases such as sclerosis tend to be Particularly, the advantages and disadvantage of these systems
fallen more and have more serious consequences [6]. Fall not will be investigated carefully. We also present challenges,
only causes physical injuries but also mental injuries. A person opportunities and trends in IoT fall detection systems. Last
used to fall often fears of falling [7]. As a result, it might but not least, we proceed a survey of 50 participants about
cause depression, losing confidence, and limit daily activities IoT fall detection systems. The survey is to build the future
[8], [9]. The fall consequences can be mitigated if a fall is IoT fall detection systems that satisfy medical requirements
treated early enough. Nevertheless, less than a half of the fall and expectations of users.
cases is reported officially. The rest of the paper is organized as follows: Section II
One of the widely used approaches to deal with fall cases discusses an overview of IoT fall detection systems. Section
and their consequences is to apply fall detection systems that III provides classification and comparison of IoT fall detection
can detect and inform fall cases in real-time for responsible systems. Section IV discusses challenges and trend of IoT fall
persons such as caregivers or relatives of elderly people. In detection systems. Finally, Section V concludes the work.
 2

II. M OTIVATIONS - OTHER SURVEY AND REVIEW PAPER OF

I OT FALL DETECTION SYSTEMS
Minh will cover this Due to the rapid improvement of
sensors and related analytical approaches, there is a need for
frequent re-evaluation of recent trends and and direction of
development.

III. OVERVIEW AND CLASSIFICATION OF I OT FALL

DETECTION SYSTEMS
need some figures In general, an IoT fall detection system
architecture consists of three main layers including sensing
layer, gateway layer, and cloud layer. Sensing layer consists
of devices or a group of devices responsible for acquiring Fig. 1: Acceleration of a person when fall occurs
data such as bio-signals, motion-related signals, images or
contextual data. The collected data can be pre-processed and
then forwarded into the next layer such as gateway layer. motion-related data such as 3-dimensional acceleration or 3-
A gateway layer include several gateways which are mainly dimensional angular velocity or both of them. Other data such
responsible for receiving data from a sensing layer and for- as contextual data and bio-signals may be also acquired. The
warding the data to cloud layer. Depending on the application, sensed data is sent to a micro-controller via a or several
the gateway layer can consist of several sub-layers that can be data buses such as I2C or SPI. At microcontroller, data can
edge layers or fog layers. In addition to main responsibility be temporarily stored, pre-processed or fed into light-weight
of receiving and transmitting data, these layers can offer algorithms. For achieving energy efficiency, a micro-controller
many other services such as distributed data storage, data has low clock rates such as 8 Mhz, 16 Mhz or 32 Mhz, and
processing (e.g., ECG feature extraction), data compression supports several sleep modes. These clock rates are enough
and push notification. The detail information of these services for real-time applications as a wearable device does not run
is discussed in the state-of-the-art work []. These services heavy computation algorithms. Instead, heavy computation
help to improve quality of service significantly e.g., improving algorithms are mainly run at gateways that have more powerful
security levels and energy efficiency. Depending on the ap- hardware including memory and processing capability.
plication, the number of sub-layers varies and offer different At micro-controller, the data can be pre-processed data
services. with light-weight methods or kept intact depending on the
Cloud layer and end-user terminals are responsible for application. The data is then forwarded into a wireless module
offering many advanced services that deal with global data via a wire protocol such as UART, I2C, or SPI for sending
storage, big data, heavy computation, and real-time streaming to a gateway. The wireless module can be Bluetooth Low
[]. End-users such as medical caregivers can use terminals to Energy (BLE), nRF, Wi-Fi, or 6LoWPAN (IPv6 over Low
access real-time or analyzed data, receive real-tine notification -Power Wireless Personal Area Networks) depending on the
about abnormalities such as a fall case. application. Each wireless communication protocol has its own
IoT fall detection systems can be categorized into 4 types advantages and disadvantages. For example, BLE, nRF, and
including wearable-based, camera-based, wireless-based and 6LoWPAN do not support high data rate (i.e., higher than few
hybrid systems. These systems are discussed in detail as the Mbps) but their power consumption is low such as around 45-
follows: 60 mW. Wi-Fi supports higher data rate but consumes much
more power such as 200-250 mW []. Due to experiments
A. Wearable-based IoT fall detection system and previous state-of-the-art fall detection systems based on
In general, wearable-based fall detection systems rely on acceleration and angular velocity, the required data rate of
motion-related signals collected from sensors and the fall these data is not high (i.e., less than 100 samples/second in
phenomena based on these signals. For example, when a fall which each sample can be 16-24bps) []. Therefore, low power
occurs, body acceleration often fluctuates and reaches peaks wireless protocols such as BLE and nRF are more preferable
shown in Fig. ??. for saving energy consumption.
Wearable-based IoT fall detection systems can be classified In many IoT fall detection systems, a fall detection algo-
into two types. The first type mainly used for indoor consist rithm is often run at gateway for detecting fall cases as gate-
wearable devices that are often light-weight, small and energy way can successfully run complex algorithms e.g., machine
efficient. Wearable devices can be often placed at wrist, ankle, learning and deep learning. These
chest, or belly [11], [13]–[19]. Some systems rely on a single The second type of wearable fall detection system can
wearable device while others may use several wearable devices be applied for indoor and outdoor applications. The system
placing at different part of the body such as a upper part and includes a wearable device and cloud computing. It can be
a lower part of the body. considered that a wearable device of this type is a com-
A wearable device is equipped with sensors, a micro- bination of a wearable device and a gateway mentioned in
controller and a wireless module. Sensors mainly collect the first type. The wearable device often has higher clock
 3

speeds such as 84 Mhz, 184 Mhz, and 216 Mhz and more B. Camera-based IoT fall detection systems
capability memory. A micro-controller of the wearable device Videos have been used as a viable technology for non-
can process data with more complex algorithms. The wearable invasive fall detection mechanism. The system is composed of
device is equipped with 4G/LTE/5G and sends the processed various cameras that streams videos to a dedicated hardware.
data into cloud server for further processing. The third type is The work of [24] used Microsoft Kinetic and On Ground
the smart-phone or smart-watch based fall detection system. Event Segmentation followed by On Ground Event Features
Acceleration and angular velocity are the main signals used for fall detection of elderly people. To be efficient, the system
for detecting fall. Smartwatch or smartphone can be used as a needs the deployment of multiple sensors inside the house.
wearable device for fall detection and classification. Casilari Additionally, the system accuracy might be degraded by the
et al. uses a smart watch and a smart phone for detecting fall. sunlight. A depth camera combined with algorithms for track-
The smart watch uses accelerometer and gyroscope to collect ing human joints has been elaborated in [25]. The system
3-D acceleration and 3-D angular velocity. The collected data works in normal and in a low-light conditions. The support
is transmitted via BLE from the smart watch to the smart vector machine, SVM, has been used for fall detection which
phone that acts as a gateway to process data and detect has an accuracy of 97.6%. Recently, a 3D-CNN (convolution
a fall case. When a fall case is detected, the smart phone neural network) has been elaborated for fall detection and hu-
sends a push notification message via 3G/4G to cloud servers man activities classification using spatio-temporal information.
[20]. Kolodziej et al. proposes a fall detection system using The work of [26] used IR camera along with a 3D-CNN
a smartphone. A smart-phone collects acceleration, angular model to detect six types of human activities among them
velocity, and magnetic field strength signals. The collected is falling. A three-stream convolution neural network (CNN)
data is processed to extract useful information including fall based fall detection system using multiple cameras has been
detection and classification [21]. Tsinganos et al. proposes been investigated in [27]. The CNN infers a decision based
a smartphone-based fall detection system for elderly people. on the human silhouette, motion history image, and dynamic
The system introduces an advanced algorithm using motion- image.
related data to detect fall and classify activities of daily living The issues associated with video based fall detection meth-
[22]. The algorithm are based on the Continuous Wavelet ods are the need to have cameras installed in every corners
Transform, threshold and a k Nearest Neighbour classifier of the house and the computational cost associated with
to achieve the target. The proposed algorithm achieves high the inference using deep-learning algorithms. A promising
fall detection accuracy up to 97.53% sensitivity and 94.89% approach is to combine RGB cameras and wearable sensors for
specificity. Dogan et al. introduces a two-step fall detection fall detection [28]. In [], the authors present a vision-baed fall
system using smartphone sensors. The system identifies and detection system using attention guided bi-directional LSTM.
distinguishes fall types including fall from standing, fall from The proposed approach first uses Mask R-CNN to extract
sitting, fall from walking, fall from jogging, and fall from lying moving objects from the noise background. Then, Convolution
[23]. The proposed fall detection approach can achieve up to neural network is used for extracting features. Finally, Bi-
the accuracy of 95.65 %. directional LSTM is applied for detecting fall events. The
proposed method is verified via both public and self-made
dataset. The proposed approach can reach to an accuracy of
TABLE I: Wearable-based fall detection systems 96.7 %
Tab. II summarizes the recently published fall-detection
Work Data Algorithm Acc(%) system using video technology.
[13] 3D acceleration 2-level threshold for fall –
peak and impact peak TABLE II: Video-based systems for fall detection
[14] 3D acceleration 2-level threshold for impact –
peak Work Technology Algorithm Accuracy
[11] 3D acceleration, 3D 3-stage algorithm with sev- – [24] Microsoft Kinetic three-stage algorithm: characterization –
angular velocity eral thresholds segmentation, and feature extraction
[15] 3D acceleration machine-learning 99.92 [25] depth camera SVM 97.6
[16] 3D acceleration [] depth camera bi-directional LSTM, CNN 96.2
and time series 2-level threshold 96-97 [26] IR camera 3D-CNN
[17] 3D acceleration support vector machine, k- 98.48 [27] multiple cameras 3D-CNN
nearest neighbors [28] RGB camera 4 ML algorithms: RF
[18] 3D acceleration Ensemble Bagged Tree 99.09 plus wearable sensors SVM, MLS, KNN
[19] 3D acceleration, Gaussian mixture models 90.25
heart rate
[22] 3D acceleration threshold and k-nearest 95-97.5
neighbors C. Wireless-based IoT fall detection system
[23] 3D acceleration, 3D support vector machine,k- 91-95.6
angular velocity nearest neighbors The wireless-based IoT fall detection system often targets
[21] 3D acceleration, 3D Discrete wavelet transform 96 indoor fall detection as the system relies on wireless signals
angular velocity, 3D
magnetometer
generated by its gateways or third-party devices closed by the
gateways. The working range of the fall detection system is
short as each gateway covers geographical area around 20-40
meters. To cover the large house, several gateways or repeaters
 4

have to be used. The widely-used approach for fall detection can achieve accuracy of higher than 90%. In these system,
and human activity categorization relies on Wi-Fi channel state several types of sensors mentioned above can be use together
information (CSI). CSI presents the information of wireless to improve accuracy of the fall detection systems. For instance,
signals that are propagated from a transmitter to a receiver at the fall detection system can consist of wearable devices and
some certain carrier frequencies. Particularly, both the subcar- contextual sensors.
rier phase and amplitude information can be got from the CSI.
Wang et al. [29] analyzed radio propagation models to find out IV. M ETHODOLOGY FOR SURVEYING AND DATA ANALYSIS
the correlations between different radio signal variations and
V. D ISCUSSIONS
activities. Based on the results, the authors proposed a Wi-Fi
based fall detection system that relies on CSI without hardware A. Challenges and solutions
modification and extra environmental setup. The system can Tri will cover security issues + BC solutions
achieve 90% and 94% accuracy for fall detection when using Current technology IoT wearable-based IoT device - indoor
SVM and Random Forest algorithm, respectively. Palipana and outdoor Camera-based IoT - Wireless-based Contextual
et al. [30] present a ubiquitous fall detection system using sensor-based and hybird IoT Fog Cloud Cryptography trans-
commodity Wi-Fi devices. The system uses CSI, short-term mission Privacy
fourier transform and a sequential forward selection algorithm An open question for leveraging blockchain
to extract time-frequency features and other features that are Challenge 1: Security including authority, authentication,
resilient to environment changes. The system can achieve 93% and privacy Healthcare systems process sensitive personal
accuracy of fall detection. In [31], the authors present a Wi- information on a regular basis. The use, transmission, and
Fi-based multistage fall detection system with channel state dissemination of such information are strictly controlled by
information. The authors developed a deep learning algorithm legislation. This drives very strong privacy and security re-
to distinguish falling and sitting cases. The results show that quirements for any devices that store, process, or transmit
the system can achieve 89% detection precision with 8% of healthcare data. A comprehensive review of the privacy-
false alarm rate. In [], the authors preserving algorithms for cloud computing in e-health is
reported in [?].
TABLE III: Wireless-based fall detection systems Lightweight cryptography in e-health and body area net-
work is a relatively new topic. The work of [35] elaborated a
Work Data Algorithm Acc(%)
[29] CSI SVM and Random Forest 90-94
lightweight identity-based encryption scheme for the body area
algorithm network. The scheme uses four IBE primitives: setup, keygen,
[30] CSI short-term fourier transform 93 encrypt and decrypt. Prior to the transmission, the sensor in
and a sequential forward se-
lection algorithm
the body area network (BAN) creates a public key using a
[31] CSI deep learning 89 simple string operation (ID+date), encrypts the data and send
[32] CSI Dynamic Time Warping 96 it to the doctor. To decrypt the data, the certification authority
[33] CSI discrete wavelet 82-90 (CA) generates an on-demand private key for the public key
transform,recurrent neural
network (ID+date). However, the scheme can be compromised once
[34] CSI deep learning (CNN, 64-66 the master secret key has been revealed. The experimental
LSTM) results carried on a Tmote Sky sensor show that the memory
consumption of the IBE-light is nearly equal to the symmetric
key and better than the conventional RSA.
To secure the communication between body sensors and
D. Contextual sensor-based and hybird IoT fall detection health-care provider, the authors of [36] proposed a revo-
system cable and certificateless remote authentication scheme. The
Another important group of fall detection systems is based Tate-based pairing is used. The scheme offers the following
on contextual sensors such as microphones, pressure pads, features: anonymity, mutual authentication, session key estab-
thermopiles, microwave Doppler sensors and ultrasonic sen- lishment, non-repudiation. Furthermore, the scheme is resilient
sors. Particularly, in [?], the authors present a fall detection to key escrow, supports key-revocation and forward security.
system based on acoustic. The system consists of a circular The authors of [37] adopted a 128-bit block cipher for
microphone array that is used to gather sound in a room. The data encryption in wireless body area network. To address the
gathered sound is analyzed to find out the source generating problem associated with authentication and key distribution
the sound and detect a fall case. The system was evaluated with in a constrained environment, the authors of [38] proposed a
simulated fall and non fall cases and the results showed that lightweight authentication protocol and a secure key exchange
the system can achieve 100% sensitivity and 97% specificity algorithm. The protocol uses both AES-CCM and pairing-
when using the dataset of 120 fall and 120 non fall cases. In based cryptography (certificateless public key infrastructure).
[?], [?], the authors present fall detection systems based on For example, this is not good enough and need BC-based
ultrasonic sensors. The systems consist of sensor nodes that solutions
can be built from a micro-controller (e.g., Arduino Uno and Challenge 2: Network connection and noise needs to be
Mega) and ultrasonic sensors that can be placed on the table, considered properly as they significantly impact on accuracy of
wall or ceiling. The results show that the proposed systems IoT fall detection systems. There are different types of noises
 5

such as general noise (50/60Hz noise), noise from movement • Trend 3: Combine fall detection and other services
artifacts, noise from wireless generating sources. Each type We have
of fall detection systems is more affected by a specific type
of noise. For instance, wearable-based fall detection systems
are more affected with noise from movement artifacts whilst VI. C ONCLUSIONS
a camera-based fall detection system seems not to be affected VII. ACKNOWLEDGEMENTS
with the noise from movement artifacts. Distance, temperature
R EFERENCES
and humidity affect on RSSI amplitude and CSI’s amplitude
and phase that are the main parameters of many wireless- [1] “Falls.” Updated: Jan. 2020, Accessed: Jan. 2020,
https://www.who.int/news-room/fact-sheets/detail/falls.
based fall detection algorithms whilst these are less influent [2] D. A. Sterling, J. A. O’Connor, and J. Bonadies, “Geriatric falls: injury
on wearable-based fall detection systems. In general, noise severity is high and disproportionate to mechanism,” Journal of Trauma
can be eliminated at sensor layer, edge layer or gateway layer and Acute Care Surgery, vol. 50, no. 1, pp. 116–119, 2001.
depending on the system. In many systems, some levels of [3] Stevens, J. A. and Corso, P. S. and Finkelstein, E. A. and Miller, T.
R., “The costs of fatal and non-fatal falls among older adults,” Injury
noise can be eliminated by light-weight algorithms at sensor prevention, vol. 12, no. 5, pp. 290–295, 2006.
devices (e.g., moving average filter or band pass filter) while [4] B. H. Alexander, F. P. Rivara, and M. E. Wolf, “The cost and frequency
some noises are removed by complex algorithms (e.g., wavelet of hospitalization for fall-related injuries in older adults.,” American
journal of public health, vol. 82, no. 7, pp. 1020–1023, 1992.
transform or AI-based approaches) at gateways [11]. By [5] “Important facts about falls.” Updated: Jan. 2017, Accessed Feb 2017,
switching heavy computation from sensor devices to gateways, https://www.cdc.gov/homeandrecreationalsafety/falls/adultfalls.html.
the low latency, high accuracy, and a high level of energy [6] C. Mosquera-Lopez et al., “Automated detection of real-world falls:
modeled from people with multiple sclerosis,” IEEE journal of biomed-
efficiency can be achieved [11], [39]. Although there are many ical and health informatics, 2020.
approaches for dealing with noise, it is challenging to remove [7] S. M. Friedman et al., “Falls and fear of falling: which comes first?
all noises as there are many non-detected or unknown noise A longitudinal prediction model suggests strategies for primary and
secondary prevention,” Journal of the American Geriatrics Society,
sources. vol. 50, no. 8, pp. 1329–1335, 2002.
It is necessary to have such adaptive and smart noise [8] A. C. Scheffer et al., “Fear of falling: measurement strategy, prevalence,
awareness approaches risk factors and consequences among older persons,” Age and ageing,
vol. 37, no. 1, pp. 19–24, 2008.
Challenge 3: Metrics such as accuracy, precision, sensitivity [9] R. Igual et al., “Challenges, issues and trends in fall detection systems,”
and specificity need to be considered when evaluating a fall Biomedical engineering online, vol. 12, no. 1, p. 1, 2013.
detection system. These metrics are shown in the following [10] T. N. Gia et al., “Fog computing approach for mobility support in
internet-of-things systems,” IEEE Access, vol. 6, pp. 36064–36082,
equations: 2018.
T P +T N TP [11] T. N. Gia et al., “Energy efficient fog-assisted iot system for monitor-
Accuracy = T P +F P +T N +F N ; P recision = T P +F P ; ing diabetic patients with cardiovascular disease,” Future Generation
Computer Systems, vol. 93, pp. 198–211, 2019.
TP TN
Sensitivity = T P +F N ; Specif icity = T N +F P ; [12] M. Mubashir et al., “A survey on fall detection: Principles and ap-
proaches,” Neurocomputing, vol. 100, pp. 144–152, 2013.
[13] I. Tcarenko et al., “Energy-efficient iot-enabled fall detection system
T P (T rue P ositive); T N (T rue N egative) with messenger-based notification,” in International Conference on
F P (F alse P ositive); F N (F alse N egative) Wireless Mobile Communication and Healthcare, pp. 19–26, Springer,
2016.
Many IoT fall detection systems based on machine learning [14] T. N. Gia et al., “Iot-based fall detection system with energy efficient
and deep learning such as CNN and LSTM have achieved good sensor nodes,” in 2016 IEEE Nordic Circuits and Systems Conference
(NORCAS), pp. 1–6, IEEE, 2016.
results e.g., higher than 90% in terms of accuracy, precision, [15] M. Saleh and R. L. B. Jeannès, “Elderly fall detection using wearable
sensitivity and specificity []. In these systems, results not only sensors: A low cost highly accurate algorithm,” IEEE Sensors Journal,
depend on the deep learning model but also rely on dataset vol. 19, no. 8, pp. 3156–3164, 2019.
[16] M. Zitouni et al., “Design of a smart sole with advanced fall detection
that has been used for training the model. Therefore, it is not algorithm,” Journal of Sensor Technology, vol. 9, no. 04, p. 71, 2019.
guaranteed that the proposed fall detection systems can provide [17] T. Althobaiti, S. Katsigiannis, and N. Ramzan, “Triaxial accelerometer-
a high level of accuracy for when the number of approaches based falls and activities of daily life detection using machine learning,”
Sensors, vol. 20, no. 13, p. 3777, 2020.
can provide for particular dataset when [18] A. Chelli and M. Pätzold, “A machine learning approach for fall
of fall detection systems are important parameters of a fall detection and daily living activity recognition,” IEEE Access, vol. 7,
detection system. If these parameters are not high, it infers that pp. 38670–38687, 2019.
[19] Y.-H. Nho, J. G. Lim, and D.-S. Kwon, “Cluster-analysis-based user-
A system with low values of these parameters these values are adaptive fall detection using fusion of heart rate sensor and accelerome-
not high enough low accurate system can cause false alarm ter in a wearable device,” IEEE Access, vol. 8, pp. 40389–40401, 2020.
that directly or indirectly impacts on the quality of services. [20] E. Casilari and M. A. Oviedo-Jiménez, “Automatic fall detection system
based on the combined use of a smartphone and a smartwatch,” PloS
The accuracy of IoT fall detection systems varies but many of one, vol. 10, no. 11, p. e0140929, 2015.
them can achieve a high level of accuracy higher than 90% []. [21] M. Kolodziej et al., “Fall detection using a smartphone,” in 2020 IEEE
Challenge 4: Emotion, feeling and confortability of users It 21st International Conference on Computational Problems of Electrical
Engineering (CPEE), pp. 1–4, IEEE, 2020.
is challenging to detect we can discuss about survey questions [22] P. Tsinganos and A. Skodras, “A smartphone-based fall detection system
and then we make a online survey. It is good if we can collect for the elderly,” in Proceedings of the 10th International Symposium on
100 answers Image and Signal Processing and Analysis, pp. 53–58, IEEE, 2017.
[23] J. C. Dogan and M. S. Hossain, “A novel two-step fall detection method
• Trend 1: stretchable wearable sensor using smartphone sensors,” in 2019 IEEE International Conference on
• Trend 2: signals based system Smart Computing (SMARTCOMP), pp. 434–438, IEEE, 2019.
 6

[24] E. E. Stone and M. Skubic, “Fall detection in homes of older adults

using the microsoft kinect,” IEEE Journal of Biomedical and Health
Informatics, vol. 19, no. 1, pp. 290–301, 2015.
[25] Z. Bian, J. Hou, L. Chau, and N. Magnenat-Thalmann, “Fall detection
based on body part tracking using a depth camera,” IEEE Journal of
Biomedical and Health Informatics, vol. 19, no. 2, pp. 430–439, 2015.
[26] A. Akula, A. K. Shah, and R. Ghosh, “Deep learning approach for hu-
man action recognition in infrared images.,” Cognitive Systems Research,
vol. 50, pp. 146 – 154, 2018.
[27] Y. Kong, J. Huang, S. Huang, Z. Wei, and S. Wang, “Learning spatiotem-
poral representations for human fall detection in surveillance video,”
Journal of Visual Communication and Image Representation, vol. 59,
pp. 215 – 230, 2019.
[28] H. Ponce, L. Martı́nez-Villaseñor, and J. Nuñez-Martı́nez, “Sensor
location analysis and minimal deployment for fall detection system,”
IEEE Access, vol. 8, pp. 166678–166691, 2020.
[29] Y. Wang, K. Wu, and L. M. Ni, “Wifall: Device-free fall detection by
wireless networks,” IEEE Transactions on Mobile Computing, vol. 16,
no. 2, pp. 581–594, 2016.
[30] S. Palipana, D. Rojas, P. Agrawal, and D. Pesch, “Falldefi: Ubiquitous
fall detection using commodity wi-fi devices,” Proceedings of the ACM
on Interactive, Mobile, Wearable and Ubiquitous Technologies, vol. 1,
no. 4, pp. 1–25, 2018.
[31] X. Yang, F. Xiong, Y. Shao, and Q. Niu, “Wmfall: Wifi-based multistage
fall detection with channel state information,” International Journal of
Distributed Sensor Networks, vol. 14, no. 10, p. 1550147718805718,
2018.
[32] Y. Hu, F. Zhang, C. Wu, B. Wang, and K. R. Liu, “A wifi-based
passive fall detection system,” in ICASSP 2020-2020 IEEE International
Conference on Acoustics, Speech and Signal Processing (ICASSP),
pp. 1723–1727, IEEE, 2020.
[33] J. Ding and Y. Wang, “A wifi-based smart home fall detection system
using recurrent neural network,” IEEE Transactions on Consumer Elec-
tronics, vol. 66, no. 4, pp. 308–317, 2020.
[34] T.-D. H. Nguyen and H.-N. H. Nguyen, “Towards a robust wifi-based
fall detection with adversarial data augmentation,” in 2020 54th Annual
Conference on Information Sciences and Systems (CISS), pp. 1–6, IEEE,
2020.
[35] C. C. Tan, H. Wang, S. Zhong, and Q. Li, “Ibe-lite: A lightweight
identity-based cryptography for body sensor networks,” IEEE Transac-
tions on Information Technology in Biomedicine, vol. 13, no. 6, pp. 926–
932, 2009.
[36] H. Xiong and Z. Qin, “Revocable and scalable certificateless remote au-
thentication protocol with anonymity for wireless body area networks,”
IEEE Transactions on Information Forensics and Security, vol. 10, no. 7,
pp. 1442–1455, 2015.
[37] A. Z. Alshamsi, E. S. Barka, and M. A. Serhani, “Lightweight encryption
algorithm in wireless body area network for e-health monitoring,”
in 2016 12th International Conference on Innovations in Information
Technology (IIT), pp. 1–7, 2016.
[38] D. Q. Bala, S. Maity, and S. K. Jena, “A lightweight remote user
authentication protocol for smart e-health networking environment,”
in 2017 International Conference on I-SMAC (IoT in Social, Mobile,
Analytics and Cloud) (I-SMAC), pp. 10–15, 2017.
[39] T. N. Gia et al., “Fog computing in healthcare internet of things: A case
study on ecg feature extraction,” in (CIT/IUCC/DASC/PICOM), 2015
IEEE International Conference on, pp. 356–363, IEEE, 2015.