Received May 21, 2019, accepted June 7, 2019, date of publication June 13, 2019, date of current version

June 27, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2922708

Research of Fall Detection and Fall Prevention

Technologies: A Systematic Review
LINGMEI REN AND YANJUN PENG
Department of Computer Science, Shandong University of Science and Technology, Shandong 266590, China
Corresponding author: Lingmei Ren ([email protected])
This work was supported in part by the Natural Science Foundation of Shandong Province under Grant ZR2018BF014, in part by the
Scientific Research Foundation of Shandong University of Science and Technology for Recruited Talents under Grant 2017RCJJ042,
in part by the Key Laboratory for Wisdom Mine Information Technology of Shandong Province, Shandong University of Science and
Technology, and in part by the National Natural Science Foundation of China under Grant 61502278.

ABSTRACT Falls are abnormal activity events that occur infrequently; however, they are serious health
problems among elderly individuals. With the advancements of technologies, falls have been widely studied
by scientific researchers to minimize serious consequences and negative impacts. Fall detection and fall
prevention are two strategies to tackle fall issues with a variety of sensing techniques and classifier models.
Currently, many reviews on fall-related technologies have been presented and analyzed; however, most of
them give surveys on the subfield of fall-related systems, while others are not extensive and comprehensive
reviews. In fact, the latest researches have a new trend of fusion-based methods to improve the performance of
the fall-related systems based on a combination of different sensors or classifier models. Adaptive threshold
and radio frequency-based systems are also researched and proposed recently, which are seldom mentioned
in other reviews. Therefore, a global taxonomy for current fall-related studies from four aspects, including
current literature reviews, fall detection, and prevention systems based on different sensor apparatus and
analytic algorithm, low power techniques, and sensor placements for fall-related systems are conducted in
this paper. Several research challenges and issues in the fall-related field are also discussed and analyzed.
The objective of this review paper is to conclude and provide a good position of current fall-related studies
to inspire researchers in this field.

INDEX TERMS Adaptive algorithm, classification algorithms, fall detection, fall prevention, low power
techniques, sensing techniques.

I. INTRODUCTION suffer moderate or severe health injuries, such as bruises,

Due to the declining of birth rate and the increasing in hip fractures or head trauma, etc. Falls also bring out psy-
life expectancy, population ageing has become a common chological burden, economic pressures and even impact the
problem worldwide. As reported by the World Health Orga- caregiver’s quality of life [3]–[5]. Earlier responses to the
nization (WHO), the population of elderly people aged elderly’s falls might decrease the serious consequences.
60 or over will increase to about 20 billion by 2050 from Fall detection and fall prevention are two important strate-
900 million in 2015, which accounts for 22% of the world’s gies to tackle the issue of elderly’s falls, and they have been
population [1]. However, ageing is always associated with studied over the past two decades. Especially, fall detection
decreasing functionalities, such as physical, sensory and cog- methods have been exhaustively explored by researchers.
nitive disabilities, which increase the risk of falling. It is These systems use different types of sensors to collect use-
reported that approximately 28-35% of the elderly aged 65 or ful signals for further processing and analysis, while var-
over fall each year. The risk of falling will rise as the age ious analysis algorithms are used to process the collected
increases. It is reported that elderly adults aged 70 or over data. Generally, most of the fall detection systems detect
who fall each year will increase to 32-42% [2], while they shock caused by the body impact using accelerations.Article
by Brown [6] considers only the large acceleration impact.
The associate editor coordinating the review of this manuscript and It is one of the first fall detection studies. To improve
approving it for publication was Fatih Emre Boran. the performance of fall detector, others combine with more
2169-3536
2019 IEEE. Translations and content mining are permitted for academic research only.
77702 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 7, 2019
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

FIGURE 1. Taxonomy for the study of fall-related researches.

sensors or apply complex machine learning-based process-

ing methods. For example, [7] and [8] combine a tri-axial
accelerometer and gyroscope to detect fall. Authors in [9] FIGURE 2. Taxonomy for Literature Reviews on Fall-related Systems.

proposed a multimodality fall detection, it is a three-step

detection strategy consisting of multiple signal sources, 1260 papers were collected and retrieved by the query, and
including an accelerometer, audio and images techniques. the final 150 papers were selected. This systematic search
While articles [10] and [11] use machine learning to effec- aims to provide an overview of the published papers on the
tively detect falls. Except for above research aspects, many following discussed topics. And these papers will be analyzed
other fall-related technologies have also been involved, for and concluded in the following subsections.
example, fall prevention methods, low-power technologies The rest of the article is organized as follows.
for fall detector, the selection of optimal sensor location for Section 2 gives a detailed survey on the existing review papers
high accuracy fall detection, etc. on fall-related systems. In section 3, we present compre-
Currently, many literature reviews on fall-related systems hensive reviews on fall detection and fall prevention meth-
constantly emerge, however, most of them focus on a nar- ods based on sensor apparatus and analytical approaches,
row scope or a specific category, for example, some only followed by the detailed discussion of the state of the art
consider one subfield implementation of fall-related sys- on them. The classification of low-power technologies and
tems as studies by [12]–[14], while others mainly provide the sensors placements for fall detection are concluded and
knowledge about principles, issues, trends, and challenges of presented in section 4 and section 5, respectively. We discuss
current systems as [13], [15], [16]. Generally, most of those the current research challenges of fall-related systems in
reviews are not overall overviews on fall-related researches. section 6. Section 7 concludes the paper and points out the
To address these gaps,this article examines current tech- possible future directions.
nologies and proposes a comprehensive classification on
fall-related systems, which is a category scheme with multi- II. SURVEY ON THE EXISTING REVIEW PAPERS ON
ple layers framework. Top classification layer of our proposed FALL-RELATED SYSTEM
global taxonomy is drawn mainly based on the research As mentioned above, review papers on fall-related sys-
directions of current fall-related studies, including researches tems are one of research angles of the current fall-related
of fall detection or fall prevention (pre-impact fall detec- researches for the elderly. Several review papers on
tion) algorithm, low-power technologies for fall detectors, fall-related systems have been published, which refer to
sensors placements for fall detection and literature reviews four different aspects of the fall-related systems as shown
on fall-related systems as shown in Fig. 1. Meanwhile, other in Fig. 2, including data availability perspective, low-power
layers of the taxonomy mainly exhibit the subclassifications implementation technologies [20], specific category of fall-
of each main class. Among the general category, fall detection related systems based on different technologies, such as
or fall prevention studies, which are the current hotspots, accelerometer-based [17], [18], smartphone-based [13], [19],
are explored and highlighted from two perspectives, includ- vision-based [14], multisensor fusion-based [15] systems,
ing the sensor apparatus usages and analytical approaches, etc. Of course, comprehensive surveys have also been done
respectively. This article aims to show researchers an over- by many studies. In this section, current review papers on
all classification of the existing fall-related systems and fall-related systems are surveyed systematically, which is
deliver them a good position regarding current fall-related seldom done by others.
researches.
To search latest fall-related studies for our proposed tax- A. SYSTEMATIC REVIEWS ON SPECIFIC CATEGORY OF
onomy, potentially relevant articles on fall-related researches FALL DETECTION OR/AND PREVENTION
were identified through a search in PubMed, PubMed Cen- In current fall detection/prevention studies, such as articles
tral, IEEEEXplore, Citeseer, Web of Science, and Sco- by [15] and [16], existing fall detection/prevention systems
pus with combination of keywords, including criteria: (fall are broadly categorized into three different classifications
detect* OR detect* fall* OR fall alerts OR fall risk OR fall according to the deployed sensors, including wearable
recognition OR fall monitor* OR fall predict OR fall prevent* devices, ambient sensors and vision-based sensors. Accord-
OR pre-impact fall*) OR/AND (real-world OR older* OR ingly, many previous review papers are published from
biosensor* OR review OR low-power OR energy-efficient these specific classifications, for example, review papers
OR placement). Besides, previous fall-related papers we on accelerometer, smartphone or vison-based fall detec-
have read were also included in this step. Totally, about tion/prevention systems, meanwhile, others focus on hybrid

VOLUME 7, 2019 77703

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

multisensor fusion-based fall detection methods. In the fol- Smartphones always have built-in sensors such as 3-axis
lowing, each specific category is discussed respectively. accelerometers, gyroscopes and magnetic, high performance
Generally, accelerometer-based fall detection is mostly microprocessors, etc. which make smartphone a very good
studied by many researchers, this is similar to the reviews on platform to detect or prevent a human fall. Habib et al. [13]
accelerometer-based systems. For example, Bagala et al. [17] reported a comprehensive survey on smartphone-based fall
presented an overall evaluation of accelerometer-based fall detection and fall prevention. They gave three taxonomies
detection algorithms. Totally, 13 different published algo- from three different operation phases of the system: sense,
rithms were implemented and compared by analyzing dif- analyze and communicate. For example, they illustrated a
ferent parameters, thresholds, and the phases of a fall, such taxonomy of smartphone-based fall detection and prevention
as beginning of the fall, falling velocity, fall impact and systems based on their sensing mechanism and placement of
orientation after the fall, based on a database of real-world sensor, which included context-aware and body worn-based
falls. Preliminary results showed the performances of those systems. They also classified them into threshold-based and
algorithms tested using real-world fall data were much lower machine learning-based fall detection and prevention algo-
than they were tested under simulated environment, which rithms on the basis algorithms of the analysis phase. Fur-
definitely indicates it is important to evaluate the proposed thermore, a detail comparative analysis was made based on
fall detection system in real-world conditions. Aziz et al. [21] their functional and architectural properties and quantitative
made an accuracy comparation of 10 accelerometer-based features. Finally, the article presented some challenges that
fall detection algorithms, 5 of which used threshold-based limited the performance and the usability of the systems.
methods while the other 5 used machine learning-based Casilari et al. [23] presented a thorough review and com-
approaches to detect falls. These threshold-based methods parison on Android-based fall detection systems considering
detected falls using input features of the minimum value different criteria such as the system architecture, the used
of the peak Vector Sum (VS) or/and the smallest value of sensors, the detection algorithm or response time to a fall
the VS, while five machine learning algorithms, including event. However, to get an accurate fall detection decision,
logistic regression, Naïve Bayes, Nearest Neighbor, Decision there were some challenges to be considered, such as effective
Tree and Support Vector Machine (SVM) were trained with evaluation methods and a reference framework, the actual
input features of the means and variances of the X, Y, and Z applicability of Android devices, for example, the sampling
accelerations acquired from the waist-mounted accelerome- frequency setting of the built-in sensors, position place-
ter. The final comparison revealed machine learning-based ment of the Android devices and the limited battery, and
fall detection methods have better overall performance than etc. Comparing with [13], [23] made an overall survey on
threshold-based algorithms. Meanwhile, SVM-based fall both smartphone-based fall detection and prevention sys-
detection was proved to provide the highest combination tems, which was a much more comprehensive taxonomy for
of sensitivity and specificity. Article by Bourke et al. [18] smartphone-based systems.
presented the performance evaluation of 21 existing proposed Currently, cameras are increasingly installed in subject’s
fall detection approaches with varying degrees of complex- home, which can be easily used to detect falls. And they
ity for a waist-mounted accelerometer-based system. Three are usually classified as vision-based systems. Accordingly,
different features were extracted, including velocity, impact some literature reviews focusing on this type of methods have
and posture to train the classifiers, while they were tested been studied and published. For example, Zhang et al. [14]
based on a comprehensive dataset containing normal activi- presented a comprehensive review on fall detection systems
ties of daily living (ADL), simulated falls and also continuous and algorithms aiming at automatically detecting a human fall
unscripted ADL. Different combination of impact, posture from vision-based perspective. In this article, vision-based
and velocity were tested. Results showed the algorithm with fall detection approaches were classified into three differ-
‘‘impact+posture+velocity’’ achieved 100% sensitivity and ent categories, including single RGB camera-based, mul-
specificity, and was the most suitable fall detection method. tiple camera-based and depth camera-based fall detection
Pierry et al. [22] also made a survey and evaluation of differ- methods. Single RGB camera-based fall detection methods
ent fall detection methods, which were grouped specifically commonly use shape related features, inactivity detection and
based on accelerometers, including only acceleration-based human motion analysis as the clues for fall detecting. For
methods, acceleration combined with other sensors-based multiple RGB cameras-based fall detections, the calibration
methods, and non-acceleration-based methods. Experiment of the system is essential to reconstruct the object. How-
results showed fall detection with accelerometer had best ever, this operation is a time-consuming process. For depth
performance, as it collected vital signals in generating accu- camera-based fall detection methods, the distance from the
rate analysis. Obviously, above studies mainly made per- top of the person to the floor, the last frames in a coordinate
formance evaluation of accelerometer-based fall detection and shape characterization are common features for such
systems, however, most of studies used simulated falls to test detection approaches. Furthermore, five publicly fall datasets
the related algorithms, while only [17] evaluated algorithms available were introduced in this study to provide researchers
based on real-world fall data, which is essential for the algo- useful benchmark datasets for public algorithm verification,
rithm evaluation. which is important for fair comparison of different methods.

77704 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

Combination of different data sources processed by a mul- which had two high levels of sufficient training data for
tisensor fusion algorithm can potentially improve the per- falls and insufficient or no training data for falls. For each
formance of a fall detection system. As presented in [15], taxonomy, fall detection systems were implemented with
authors concluded the challenges and issues of multisenosr different classification methods, which were investigated and
fusion-based fall detection. They categorized and described discussed detailedly by authors. Moreover, they also pointed
current multisensor fusion algorithms into three generations, out the method of treating a fall as an abnormal activity to be a
including wearable sensors fusion, context-aware sensors plausible direction, while personalized fall detection solution
fusion and wearable/ambient sensors fusion. Each catego- in supervised classification of falls is hard.
rization was described in detail. However, cost efficiency,
conflicting output, data correlation processing framework and C. COMPREHENSIVE LITERATURE REVIEWS ON FALL
computational power were challenges that should be analyzed DETECTION OR/AND FALL PREVENTION
and taken into consideration. Currently, many literature reviews also have been done from
fall detection or/and fall prevention perspectives. However,
B. LITERATURE REVIEWS ON FALL-RELATED SYSTEM most of them still lack of a comprehensive classification.
FROM ENERGY-EFFICIENCY AND DATA AVAILABILITY In this subsection, we analyzed and concluded current review
PERSPECTIVE papers from three aspects, including fall detection, fall pre-
As shown in our previous work [24], energy consumption of vention, as well as both fall detection and prevention angles.
a fall detector dramatically impacts whether users use the From fall detection aspect, Mubashir et al. [16] made
system, and further affects the quality of the services. High a comprehensive survey on fall detection systems and
energy consumption implies frequent battery recharging or the related algorithms, which were mainly divided into
replacement, which is disastrous for the elderly person to three parts, including wearable-based, ambience-based and
do this. Energy-efficiency technologies can help to alleviate vision-based fall detection approaches. Each category was
this issue and extend the use time of fall detection system. further distinguished according to the detail technology it
Wang et al. [20] made a comprehensive review on low-power referred to, for example, wearable-based methods further
technologies for wearable telecare and telehealth, which were included posture and motion-based devices. They also sum-
concluded and classified into two classes, including hardware marized and discussed the properties of these approaches.
and firmware-based methods. For each category, the arti- Finally, a brief comparison of different approaches was made,
cle detailedly elaborated recent developments and researches and they found a robust fall detection system should simul-
about how to realize these approaches. For low-power hard- taneously consider both sensitivity and specificity for prac-
ware designs, they mainly focused on the selection of appro- tical application. Existing issues of current fall detection
priate hardware, such as appropriate electronic components researches are concluded, for example, ethical issues of con-
or hardware framework, while low-power firmware methods fidentiality and privacy, the dependency risk on the individual
employed event-driven, duty cycle, feature selection or sensor body, adequate, overall and public fall dataset of real-world
selection to reduce the energy efficiency of the system. Com- for experiment and evaluation. Mohamed et al. [27] also
monly, an energy-efficient system usually combines multiple classified fall detection algorithms into three categories,
low-power technologies mentioned above to optimize and and summarized the merits and demerits of each classifica-
prolong the battery life of the whole system. tion approach, however, authors considered the video-based
From the data availability aspect, obviously, falls are infre- method was the best choice due to its versatility. The article
quent and diverse subject events, which seldom and random by Chaudhuri et al. [28] gave a systematic assessment on the
occur during the elderly’s daily life. Therefore, there lacks current fall detection systems, which contained 57 recorded
sufficient real-world fall data, especially, during fall detection papers using wearable systems and 35 non-wearable systems.
study, it is a fatal issue when training the classifiers or setting In this paper, authors analyzed and categorized fall detection
the thresholds. Though major published studies have been studies from the aspect of system evaluation, which was dif-
verified by laboratory simulations, the performances of fall ferent from the former survey papers. Results of comparison
detection algorithms are greatly affected when they are tested showed only 7.1% wearable-based systems monitored the
with real-world as clearly stated in [17] and [25]. Broadley elderly in real-world environment during evaluation proce-
et al. [25] evaluated current fall detection algorithms from dures, while no non-wearable-based fall detection studies
real-world data perspective. Total twenty-two fall detection used the elderly as subjects in a lab or a real-word setting.
articles that had been tested using real-world data in their Obviously, latest research hotspots, scuh as adaptive, Radio
research were recorded and discussed involving the data Frequency (RF)-based and adaptive-based detection methods
collection and preparation, data processing methods and the are seldom mentioned in above mentioned reviews. Recent
criterion of the performance measures. Examining results public surveys on fall detection have also been published,
showed the performance of these approaches were inconsis- such as publications by Lapierre et al. [29] and Vallabh
tent and the number of real-world falls was commonly small. et al. [30]. Lapierre et al. in [29] provided a general classi-
Khan et al. [26] presented a classification for the current fall fication on fall detection. In this article, 118 studies were
detection studies from the availability of fall data aspect, included and analyzed from the characteristics of the applied

VOLUME 7, 2019 77705

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

technologies, including the types of sensors (based on which risk diagnosis should be proposed, which reconsiders the
systems were classified into wearable-based, ambient-based functional evaluation and user experience. Oladele et al. [34]
as well as combination of wearable and ambient technique), also presented a comprehensive literature survey on fall pre-
common algorithms to detect fall, the evaluation standards vention, which aimed to clarify the usefulness of information
and outcomes. In discussion section, they put forward that technologies in fall prevention.
the results of these 118 articles should be reconsidered due As shown in [35], fall-related technologies can be split into
to the simulated conditions and the lack of the elderly adults fall detection and fall prevention. Both of them apply differ-
for evaluation. Similarly, Vallabh et al. [30] also divided fall ent types of sensors to sense related signals for further fall
detection systems into wearable-based, ambient-based, and detecting or preventing. However, reviews on fall detection or
camera-based systems. For each classification, they made a prevention discussed above only investigate one special fall
systematic analysis and conclusion on the most recent system technologies for the elderly. Currently, many comprehensive
implementations. Meanwhile, the disadvantages or limita- reviews consider both fall detection and fall prevention. For
tions of them were also discussed. Finally, they considered example, Delahoz and Labrador [36] presented the state of
and proposed personalized fall detection approach to be the the art in fall detection and fall prevention systems. They con-
trend to create high accuracy and adaptive to new human cluded the general model of both systems, and a three-level
activities, for which the lack of real fall data of the elder would classification of falling risk factors. They also made a thor-
not be an issue. oughly review and comparison of current fall detection and
Unlike fall detection technology, fall prevention (also fall prevention approaches. Hawley et al. [37] in 2014 pro-
called pre-impact fall detection) has been identified as an vided a systematic review on fall detection, prevention and
effective strategy [31] to prevent falls, which can activate monitoring of the elderly adults. They considered intrinsic
the configured fall prevention apparatus to reduce physi- factors such as independence, requirements for safety to be
cal injuries. Hu and Qu [31] conducted a systematic survey important motivation to use such technique, while extrinsic
on pre-impact fall detection technologies. Authors analyzed factors including usability, costs etc. were also important to
and discussed current pre-impact researches from multi- support users’ attitudes. The article by Chaccour et al. [35] is
ple aspects, including detection apparatus, indicators, algo- another comprehensive review on both fall detection and fall
rithms, types of falls for evaluation, and also the performance prevention system. In this study, both fall detection and pre-
of pre-impact fall detection. They also reported the limitation vention systems can be divided based on sensor technologies
of current technologies, appropriate selection of fall detection into a 4-level classification, the top layer of which included
indicators, as well as the lack of real fall data were three exist- wearable-based, non-wearable, and fusion or hybrid-based
ing limitations of current pre-impact fall detection systems. systems (new category added in this taxonomy). Authors gave
Study work by [32] mainly presented a comprehensive search a detail discussion on fall detection and prevention systems
on sensor technologies for fall prevention among institution- with respect to the sensors deployment, data processing and
alized geriatric patients. Four specific issues including fall analytical approaches. Based on their analysis, fall detec-
prevention interventions, effectiveness of fall prevention sys- tion/prevention researches should consider exploiting gait
tems on fall rate, false alarms rate and also user’s experiences, and balance assessment to provide reliable solutions for elder
such as feedbacks or possible alerts to the patients and nurse, adults.
were addressed and discussed detailedly. Results showed Obviously, most previously mentioned surveys mainly
there was no evidence that current sensor technologies to emphasize on the analysis of fall detection or/and preven-
prevent falls of people in indoor care environment would tion systems from a specific research aspect, even some
reduce fall rates. Only one study among the final selected make comprehensive surveys, they miss current significantly
12 articles in this paper reported false alarm rate up to 16%, novel categories, such RF-based, fusion-based, personal-
however, this rate was too high. Therefore, effective detection ized/adaptive fall detection methods, each of which can be
methods should be focused to make intervention success- considered as a category as the proposed taxonomy in this
fully. Sun and Sosnoff in [33] also made a survey to assess paper. In this paper, an overall review on both fall detection
the current sensing technologies that were used for fall risk and prevention is done based on recent studies from sensor
assessment in elderly people. They extracted and analyzed deployments and their analytical algorithms.
the sensing techniques used for fall assessment, information
about fallers and types of fall, the extracted features, and III. GENERAL CLASSIFICATION OF FALL DETECTION/
also fall prediction models of all selected 22 studies. Results PREVENTION
indicated that sensing technologies of inertial sensors, cam- According to our proposed top layer taxonomy for the stud-
eras, pressure sensors and laser sensors were four potential ies of fall-related researches, study of fall detection or/and
approaches for high accuracy fall risk assessment in older fall prevention is another research hotspots, which has been
adults, however, due to the variation in signal measurements, exhaustively explored by current researchers in recent years.
parameters selection and modelling methods, a diverse range However, as it is shown in above section, current literature
of diagnostic accuracy was reported when they were used reviews on fall detection and prevention systems mainly
for activity assessments. Therefore, a clinical meaningful fall focus on a narrow scope but lacking of a comprehensive

77706 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

orientation or inclination changes, which can be sensed and

measured by Micro Electro-Mechanical devices, such as
accelerometers, gyroscopes, or other types of sensors like
barometers or magnetometers, etc. Normally, one or more
above sensors are placed on different body parts of the elderly
to measure the abrupt changes of human body, which are fur-
ther used to detect or identity falls from normal ADL. Except
for above reasons, features of miniaturization, portability,
low-cost and real-time make inertial sensor(s)-based systems
to be potential and popular devices among the elderly for
practical services. Majority of current fall detection and fall
prevention researches apply inertial sensors to collect body
information so as to detect or prevent fall events. 2-/3- axial
accelerometers or/and gyroscopes are commonly used sen-
FIGURE 3. Taxonomy for fall detection and fall prevention from sensor sors in most studies, other inertial sensor apparatuses are also
apparatus aspect.
used in some studies to increase the accuracy of fall detection
or prevention. Current studies on inertial sensor-based fall
classification. Additionally, more novel approaches have
detection and fall prevention systems are summarized and
appeared after those mentioned reviews were published,
concluded in our paper as shown in Table 1.
which can be considered as new categories, for example,
Listed inertial sensors no matter in custom-built
RF-based methods, sensor fusion-based methods, and per-
devices or general development board, or even in smart-
sonalized/ adaptive methods. Therefore, a further comple-
phone/wristwatch are used to sense the motion changes
mentary taxonomy on fall detection and fall prevention is
of the monitored body to detect the abrupt changes, ana-
proposed in this paper by extending the above classification
lyze gait, monitor the body orientation or/and assess the
for the study of fall-related researches. Especially, we pro-
muscle control signals, while they are always tied to the
pose two global classification schemes on fall detection and
body. When a fall occurs, body changes can be sensed and
prevention from the sensor apparatus and analytical algorithm
used to detect a fall. Many fall detection studies use single
standpoints, respectively.
type of wearable inertial sensors to achieve, for example,
A. SENSOR APPARATUS-BASED CLASSIFICATION SCHEME Shahzad et al. [38] achieved an accelerometer-based fall
As it is shown in previous studies, fall detection system prevention system in 2017, then they further proposed a
usually detects the body impact to trigger an alarm, while high accuracy fall detection approach using only accelerom-
fall prevention system extracts the gait information to get eter. Gyroscope was also used as measuring sensor for fall
early fall alert. However, both fall detection and preven- detection or prevention as shown in Bourke and Lyons [64],
tion system always use accelerometers, gyroscopes, pressure Su et al. [65], and etc. Meanwhile, Lu et al. [68] and Chac-
sensors, video/depth camera, microphone, or/and radio fre- cour et al. [69] used only pressure sensor(s) to detect or
quency, etc. to determine falls or fall risks, in other words, prevent falls. Further, there are also only electromyography
they usually apply similar types of sensor apparatus with or inclinometer-based fall detection or prevention system.
different numbers, which is the same idea as drawn from However, among those listed inertial sensors, but not limited
current literature studies [32], but they have different final to those sensors, accelerometer is the most popular and
objectives. Therefore, in our paper, fall detection and fall widely used sensor in fall detection/prevention systems as
prevention are considered to have the same categories from shown in Table 1. As accelerometer can sense and extract
the sensor apparatus angle, which is a global classification multiple significant parameters for fall detection/prevention,
scheme for fall detection/fall prevention. and can be feasible, fast, real-time and effective solutions to
In the proposed scheme, four major classes are reported detect falls. What is more, as shown in Table 1, it is clear
and presented based on sensing apparatus used in the existing that accelerometer-based fall detection or prevention systems
fall detection and fall prevention systems as shown in Fig. 3, have high performance than those systems using pressure,
they include inertial sensor-based, context-based, RF-based electromyography, or inclinometer sensor.
and sensor fusion-based fall detection/fall prevention. Worthy Inertial sensors-based fall detection and fall prevention
mentioning, the classification drawn in Fig. 3 is totally based studies are one of the most current hotspots and trends. They
on current literature publications. In the following subsubsec- have many advantages, such as portable while easier to be
tions, each category is illustrated and explained thoroughly. implemented, few privacy issues, high accuracy, real-time
and etc. Nevertheless, there are still some limitations in such
1) INERTIAL SENSOR(s)-BASED FALL DETECTION AND FALL approaches. Firstly, elderly to be monitored are required to
PREVENTION wear such device on his/her body, which is uncomfortable
It is known that most falls are always accompanied by sud- and disastrous, as it is an intrusion for users. Second, sensor
den body changes, such as severe crash, significant body placement and external noise will affect the performance of

VOLUME 7, 2019 77707

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 1. Current studies on inertial-based fall detection and fall prevention systems.

the system. Furthermore, fall is a rare event, which leads to 2) CONTEXT-BASED FALL DETECTION AND FALL
a lack of real-world fall data to give proper thresholds or PREVENTIONS
training related classifiers, further bring out poor fall detec- As mentioned above, many review studies categorized cur-
tion/prevention performance. rent fall-related systems into wearable-based, ambient-based

77708 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 2. Current studies on context-based fall detection and fall prevention systems.

and vision-based fall detection/prevention systems. How- For ambient-based system, acoustic, vibration, and pres-
ever, in this paper, ambient-based and vision-based fall sure signals are collected to track the body within the sensor’s
detection/prevention are considered as context-based cate- view. Mel-Frequency Cepstral Coefficient (MFCC) features
gory. As both methods detect fall by sensing environment of acoustic signal can be extracted to capture the move-
information to track the movement of the body. In this cat- ments of the user, which are further used to classify fall
egory, external sensors such as microphone, pressure sensor, and ADL, such as approaches proposed by Droghini et al.
infrared sensor, camera, thermal sensor and etc. are attached in [82] and Irtaza et al. in [83]. Falls and ADLs always show
around the surrounding where the individual stays in, such different vibration patterns, which is the basic theory of such
as bedroom, bathroom or a home to detect or prevent falls. category system. Vibration signal can be collected by var-
Current studies on context-based fall detection and fall pre- ious pressure sensors (piezoresistive or resistive) [89], [91]
vention systems are listed in Table 2. or sensor pad/mat [90], [92] on the floor, which are always

VOLUME 7, 2019 77709

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

used in both fall detection/prevention system. Infrared to monitor the body behavior for fall detection/prevention
sensors have also been used in fall detection [97] and pre- without disrupting the normal life of the individual. More-
vention systems [98]. Most notably, both audio and pres- over, due to the development of technologies, the price of
sure signals are commonly used to detect falls, however, camera decreases rapidly, therefore, it becomes popular and
it is clear that audio-based approaches have better perfor- is increasing utilized in elderly’s daily life, which promote
mance than pressure-based methods. Meanwhile, both pres- camera-based fall-related system to be a plausible research
sure and infrared sensors have been applied for fall detection direction. However, there are many disadvantages in this cat-
and fall prevention, but Infrared sensors are more preferred egory. Firstly, camera-based system always applies complex
by researchers in recent years. In addition, compared with computer vision and image processing techniques to mon-
inertial-based system, ambient devices are the least intrusive itor the individual, which requires considerable computing
as they are always unobtrusive and have minimum inter- and storage capacity to run the real-time algorithm. Second,
action with the individual, which also implies few privacy acquiring real falls image/video of the individual is difficult,
and security issues. However, in this category, there are still as this always refers to privacy issue, which is a serious issue
plenty of problems making this type method not to be the for most people. Thirdly, camera(s) in systems are always
best choice. Coverage of this category is one of the most seri- fixed at fixed placements, which means limited capture space
ous difficulties. Ambient-based systems always place sensors can be monitored. What is more, for such category system,
only indoors or in one room, which brings out dead spaces installation and calibration of cameras are also difficult to
or blind spots in fall detection/prevention, in other words, operate.
these methods have limited detection range. Moreover, they
3) RF-BASED FALL DETECTION AND FALL PREVENTION
also make an assumption that only one individual stays in
Obviously, radio frequency can also be classified into
the room. Besides, ambient sensors are easier affected by
context-based system. However, in this paper, RF-based
external environment, for example, other falling materials in
methods are taken out due to the type of signal, size of
the monitoring room, floor types, and various noise, which
data, and the special sensor apparatus, meanwhile, recent
affect the performance of the system and produce many false
development and trend in this category are also considered.
alarms.
These technologies track the fluctuation of radio frequency
With the popularity of cameras in our daily life, cam-
signals or wireless channel state information (such as WiFi,
eras are embedded into fall monitoring system gradu-
Bluetooth) to detect or predict falls, as the strenuous body
ally to acquire information, which are also considered as
movement speed brings out abnormal changes on RF signals.
context-based fall detection/prevention systems. Many stud-
RF-based system can be categorized into two classes as con-
ies using capture system to tract the head trajectories, body
cluded in Table 3, including radar frequency-based and wire-
shape changes, or body posture of the monitoring subject
less channel-based system. To detect falls, Tian et al. [118]
to detect or prevent falls. These capture systems can be
analyzed signals collected from multi-antenna Frequency
RGB camera(s), depth camera(s) (Kinect), thermal sensor(s),
Modulated Continues Wave (FMCW) radio and extracted
or even multiple cameras combination. The most simple and
complex spatio and temporal features to train Convolutional
common vision-based method applies a single camera, such
Neural Networks (CNN). Tang et al. [121] proposed a FMCW
as de Miguel et al. [102] used k-Nearest Neighbours (kNN)
radar-based fall prevention system, they constantly measured
algorithm to analyze the silhouette change over time with
the distance between radar and surrounding environment
a camera module, which is the cheapest system among this
and analyzed the relationship between body motion and
category and is easy to setup, however, the performance
the radar frequency to predict falls. Wireless channel state
of such approach needs to be further improved due to the
information-based fall related system estimates fast changes
limited area coverage. Multiple cameras are used in many
in wireless signal caused by different human activities, they
studies to cover a wide detection area, for example, Fan
can be WiFi or Bluetooth. Wang et al. [124] proposed a novel
et al. [104] applied 8 cameras mounted around the room in the
fall detection system based on WiFi devices. They demon-
proposed method to improve the performance of fall detection
strated that wireless channel state information can distinguish
system. Depth information from camera(s) can increase the
fall and fall-like activities successfully. Apparently, radar
accuracy of the system. As depth camera(s), such as Kinect
frequency signal is ubiquitous, based on which automatic
can be used to calculate the distance between the person
fall detection can be achieved conveniently without user’s
and the floor to improve the performance of the system. For
involvement, that is to say, this method is nonintrusive. How-
instance, both articles of Zhao et al. [109] and Li et a. [110]
ever, RF-based technology also has coverage issue, wireless
tracked the key joints of the body using depth camera to
network is always deployed within the limited range of the
detect or prevent falls. Moreover, thermal sensors are also
house.
widespread used in fall-related researches, which have high
accuracy up to 99.7%. Compared among the above mentioned 4) SENSORS FUSION-BASED FALL DETECTION AND
detection means, depth camera-based methods attract much FALL PREVENTION
more attentions of current researchers. Obviously, camera can Preliminary studies have demonstrated single sensor-based
record the body image continuously, which can be applied fall detection system often has low accuracy or/and high false

77710 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 3. Current studies on RF-based fall detection and fall prevention systems.

alarm, it needs additional information to improve the accu-

racy of the system. Homogeneous or heterogeneous types
of sensors can be fused together to improve the perfor-
mance of fall detection/prevention systems. Current studies
on sensors fusion-based systems are listed in Table 4. For
FIGURE 4. An example general model of fall detection/prevention from
homogeneous sensors fusion, they can be similar or the same analytical algorithm aspect.
types of sensors, such as fusion of inertial sensors, fusion
of ambient sensors, etc. Fusion of inertial sensors among they also have other limitations, for example, information
triaxial accelerometer, triaxial gyroscope and triaxial mag- redundancy, robust fusion algorithm.
netometer was proposed to detect falls by de Quadros et al.
in [11]. Leone et al. [128] combined electromyography and B. ANALYTICAL ALGORITHMS-BASED CLASSIFICATION
accelerometer information to detect pre-impact falls, more SCHEME
than 750ms lead detection time was allowed before a fall Generally, smart sensors are attached to/near the body parts
happened. Certainly, different types of sensors can also be of the individual to measure significant signals, which are
combined together to exploit different data sources for fall used to distinguish falls from ADLs with some classification
detecting or preventing, which is called as heterogeneous sen- algorithm. A general example of fall detection/prevention
sors fusion method in this paper. For instance, in our former system model from analytical algorithm perspective is shown
study [9], a multimodality-based fall detection scheme based in Fig. 4. Totally, four component parts are associated with
on accelerometer, microphone and camera was proposed and this model, in which data collection and pre-processing is the
verified. For fall prevention, Rantz et al. [139] showed a com- first step of the system. It collects data from various sensors as
prehensive fall prevention approach based on heterogeneous mentioned and discussed in the above section. The collected
sensors fusion of pulse-doppler radar, a microsoft Kinect, data directly or indirectly reflect the body motions. Then
and 2 web cameras. Results demonstrated that there was sig- data pre-processing methods such as Kalman filters, mean
nificant performance improvement when system combined filters, or advanced integration method are used to remove
several types of sensors together to provide multiple data noise and external impact. In fall-related systems, distinctive
sources, as sensors fusion-base system was able to provide features are significant attributes to distinguish fall and non-
sufficient information on human activities or gait balance fall, which should be extracted from the raw data. And they
characteristics. Currently, with the recent development in play vital roles in fall detection/prevention algorithm. For
science technologies, more novel sensors spring out and can example, the most popular features of inertial-based systems
be combined together to improve the performance of the are the magnitude of the acceleration and the angular velocity.
system, which is a trend in such domain. However, few fall In context-based system, aspect ratio of image, MFCC fea-
detection or prevention devices or systems are used in real-life tures are common extracted features. After feature extraction
of the individual so far, though there are many works have and analysis, the chosen classification model is trained with
been done. This is because of the lacking of high accuracy the extracted features to detect/prevent falls. Actually, models
fall detection approaches, in other words, sensor fusion-based used in current fall detection or fall prevention systems can
methods still face low performance issue. Except for this, be further divided into three classes based on the applied

VOLUME 7, 2019 77711

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 4. Current studies on sensors fusion-based fall detection and fall prevention systems.

classification algorithm, including threshold, non-threshold

and fusion-based analytical algorithms, as shown in Fig. 5.
This classification is based on current literature publications,
too. In the following subsubsections, each category will be
illustrated and explained thoroughly.

1) THRESHOLD-BASED FALL DETECTION AND FALL

PREVENTION
Currently, most fall detection and prevention studies apply
threshold-based approaches to detect/prevent falls. Fall
events can be detected or prevented by comparing the col-
lected data with a setting reference value (threshold). How- FIGURE 5. Taxonomy for fall detection and fall prevention from analytical
algorithms perspective.
ever, threshold set in the algorithm significantly affect the
performance of the system. A high threshold value brings
out large amount of fall missing issues, while a low threshold threshold(s) and adaptive threshold(s)-based methods as con-
value causes false alarms. Therefore, appropriate thresholds cluded in Table 5.
should be chosen and set. Actually, current threshold-based Due to the low computational complexity, fixed threshold-
approaches can be further classified into two groups: fixed based method has been widely applied by current fall

77712 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 5. Current studies on threshold-based fall detection and fall prevention systems.

detection and fall prevention studies. For example, Sivaranjani et al. [143] predicted falls by analyzing data
Razum et al. [140] selected optimal threshold values for two collected from both gyroscope and tri-axial accelerometers.
features of sum vector magnitude and euler angle to dis- The collected signals of acceleration and angular velocity
tinguish falls from ADLs based on the receiver operating were analyzed and compared with fixed threshold values
characteristics (ROC) curve, which has been commonly (±3 g, 0.5 rad/s) to determine pre-fall and inflate the airbag.
applied in former studies. Two optimal thresholds were Clearly, for above discussed algorithms, pre-defined thresh-
selected and analyzed for three situations of each feature olds should be determined firstly, however, it is commonly
separately and the combination of two features, respectively. confirmed empirically in most publications. Furthermore,
The best results were obtained when sum vector magnitude as shown above, different thresholds are set in different
was set to 4.1 g, while Euler angle between the direction publications though they are used for the same detecting
of earth gravitational field and the vertical axis of sensor parameters. This is because each individual has different
was set to 70 degree. Abdelhedi et al. [51] also proposed a characteristics, such as age, sex and physique etc., which
threshold-based approach to discriminate falls from ADLs. affect the setting of the fixed thresholds. Therefore, fixed
In the proposed method, two fixed thresholds of sum vec- threshold-based fall detection system cannot perform well
tor magnitude were set to 0.6 g and 1.8 g respectively to for different individuals. In other words, fixed thresholds for
detect free fall phase and impact phase, while feature of one may be unsuitable for others to ensure high fall detection
body tilt was set to 60 degree to detect inactivity phase. performance of the system.

VOLUME 7, 2019 77713

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 6. Current studies on non-threshold-based fall detection and fall prevention systems.

Clearly, fixed threshold-based methods have low recogni- refine personalized threshold for high accuracy fall detec-
tion ability, and always result in high false alarms. Dynamic tion system. Adaptive threshold methods were also pro-
or adaptive threshold-based fall detection and prevention posed for fall prevention. In article by Otanasap [40], they
methods have been proposed to solve these problems. proposed a pre-impact fall prevention system using adap-
Wu et al. [127] proposed an adaptive threshold method to tive threshold model, which automatic adjusted threshold
detect falls based on a multivariate control chart. This adap- based on motion history of the user. Obviously, experiment
tive threshold method had high detection performance as it results concluded in Table 5 clearly show personalized or
considered individual historical data, that is to say, this con- adaptive thresholds-based fall detection/prevention systems
structed fall detection was a person-specific method. In article considering user’s characteristic and other influence fac-
by Ren and Shi [144], different user groupings, including tors have high performance than these fixed threshold-based
different gender, age, height and weight, were analyzed to systems. We believe this category is a inevitable trend for

77714 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

the widespread use and extension of the system in real method, however, it requires high computing capability and
life. storage volume in general. Recently, fusion method combin-
ing threshold or/and non-threshold method has sprung up
2) NON-THRESHOLD-BASED FALL DETECTION AND FALL to increase the accuracy of the system, which integrates the
PREVENTION advantages of the combined methods. Based on current stud-
Non-threshold-based methods always use complex algo- ies, fusion-based methods can be divided into homogeneous
rithms to distinguish or predict falls from ADLs. They mainly and heterogeneous-based approaches as listed in Table 7. For
apply machine learning algorithms or statistics process algo- homogeneous fusion-based methods, they can be multiple
rithms as concluded in Table 6. threshold algorithms voting to determine falls as proposed by
For machine learning-based methods, commonly used Poonsri et al. [154]. Combination of machine learning meth-
algorithms in fall detection or prevention system are kNN, ods can also be used to increase the accuracy of fall detection
Support Vector Machine (SVM), Naïve Bayes, Hidden system as proposed by Cheng and Jhan [155]. Meanwhile,
Markov Mode (HMM), random forest, fuzzy logic, etc. For homogeneous fusion-based method has also been applied to
instance, Yu et al. [146] developed a HMM-based fall detec- prevent falls. Su et al. [65] showed three hierarchical clas-
tion algorithm using a single 3-axial accelerometer. Raw sifiers based on Fisher discrimination analysis to predict
acceleration signals were analyzed using Gaussian distribu- falls. For heterogeneous fusion-based approach, the combi-
tions for hidden states to train HMM models. In article by nation of threshold and Non-threshold method have been put
Zhou et al. [147], three CNNs were used for fall recognition forward to distinguish or predict falls. Currently, multiple
by training three feature sets extracted from capture infor- combination strategies, such as combination of threshold and
mation of wave radar and optical cameras. The combination multiple kernel learning SVM [10], or threshold and kernel
decision of the three CNNs gave the final fall detection density estimation [48], etc. were proposed to reduce false
results. SVM is another popular machine learning technique, alarms. Comparison among Table 5, 6 and 7, it is clear
which needs to find a hyperplane to ensure the largest margin fusion-based fall detection/fall prevention system has high
between different classes. SVM is also used to distinguish performance than single threshold or non-threshold-based
falls from ADLs, as published in Min et al. [151], they method. Therefore, we also consider fusion-based method to
extracted 32 features from data collected by Kinect sensor be one of future study directions.
to train the proposed classifier. Apparently, machine learning
techniques can also be used to predict falls, for example, IV. CLASSIFICATION OF LOW-POWER TECHNOLOGIES
SVM-based preimpact fall detector was proposed by Zhen FOR FALL DETECTION
et al. [148] and Aziz et al. [152]. Neural networks were con- Since a wearable fall detector is typically powered by a bat-
structed by Steffan et al. [150] to prevent falls. Certainly, tery, energy-efficient approach is essential for such service.
there are many other machine learning-based fall detection However, traditional studies usually commit to the accuracy
and fall prevention methods which have not been mentioned of fall detection, but neglect the fact that fall detection algo-
in this paper. Obviously, from Table 5 and Table 6, it is clearly rithm always runs on a microcontroller with limited comput-
observed that the performance of machine learning-based fall ing, storage and energy resources, which decide the limited
detection system is higher than that of threshold-based system usage time of the detector.
by training classifiers with extracted features. Currently, some researches on low-power fall detector have
Statistics process can also be used to detect or prevent gradually been put forward. As we all know, the elderly
falls. Wu et al. [127] applied Hotelling’s T2 statistic to performs ADLs for most of the time, while falls seldom
detect falls, while Hu and Qu [153] proposed ARIMA-based occur. Therefore, in our former study [24], an energy efficient
statistical process-based pre-impact fall detector. Actu- scheme of using a low sampling rate during most of time,
ally, both machine learning and statistics-based fall detec- but a high sampling rate when there is a possible fall to
tion/prevention algorithms have been widely studied, as they increase the performance of fall detection algorithm was pro-
can achieve high accuracy than threshold-based fall detec- posed, which was called a segmented sampling rates scheme.
tion/prevention methods. However, they are complex comput- This proposed low-power fall detection method was verified
ing processes and commonly require high computing volume to improve both energy efficiency and detection accuracy.
compared to threshold-based approaches, which are the main Article by Wang et al. [158] also concentrated on adjusting
limitations for these methods. the sampling rate to get the goal of energy efficiency. The
proposed low-power fall detection algorithm dynamically
3) FUSION-BASED FALL DETECTION AND FALL PREVENTION adjusted the sampling rate of the sensor and managed wireless
Both threshold and non-threshold-based systems have their transmission to reduce power consumption. Testing results
advantages and disadvantages. Threshold-based method is showed the proposed energy-efficient algorithm achieved
always light-weight algorithm and is easy to be imple- slightly better than the algorithm without using low-power
mented in wearable detector, however, the performance is idea. Solaz et al. [159] presented an energy-aware fall detec-
hard to ensure. Non-threshold method can improve the perfor- tion integrated circuit, which included a Programmable Trun-
mance of the system effectively compared to threshold-based cated Multiplier (PTM). It combined the power-reduction

VOLUME 7, 2019 77715

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

TABLE 7. Current studies on fusion-based fall detection and fall prevention systems.

77716 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

benefits of the standard truncated multipliers with the ben- performance of fall detection system. 21 corresponding
efits of programmability. Yuan et al. [160] proposed an attributes were extracted for accelerometers located on the
interrupt-driven low-power fall detection algorithm based on waist, chest, thigh and ankle, and were trained by the pro-
a digital accelerometer (ADXL345). ADXL345 supports var- posed classification model for fall recognition. Chest was
ious interrupts, which were used in the proposed fall detection proved to be the best location for advanced posture recog-
algorithm to trigger different fall phases. The special capabil- nition. Meanwhile, authors also compared the accuracy of
ity of it allowed adaptive status changes of MCU between accelerometer-based fall detection and posture recognition
the deep sleeping mode and the working mode according to with four different located sensors. Study of [42] was another
interrupts, which is the key idea of power saving for the pro- study focused on the best subset of body segments for fall pre-
posed fall detector. The article by Gia et al. [161] investigated vention. An ad-hoc designed machine learning algorithm was
energy efficient fall detector in different configurations and proposed and tested with the recorded data of different sub-
operating conditions from hardware (such as the choice of sets of all body segments to determine the minimum number
micro-controller, motion sensor, and transmission module) of segments to get a good detection performance. Experiment
and software (such as the choice of sampling rate, trans- results showed the information collected by all the different
mission distances and transmission condition) perspectives, body segments is redundant for pre-impact fall detection,
and they presented the optimal hints for the implementation only the kinematics of upper and lower distal extremities
of low-power system. In article by Rahmani et al. [162], are adequate signals for high performance of pre-impact fall
authors analyzed multiple factors that can reduce the power detection.
consumption of the system, such as micro-controller, 3D Obviously, the determination of the number and appro-
accelerometer sampling rate and Bluetooth technology. Wang priate placements of the wearable sensors can improve the
et al. [163] proposed an optimal low-power fall detector that performance of fall detection and fall prevention system.
contained triaxial accelerometry and barometric pressure sen- Current related studies mainly compare previous algorithms
sor. They mainly used a combination of both hardware and with information collected by different sensors located on
firmware-based method to reduce the power consumption different body segments to decide the best sensor locations,
of the fall detector, such as the selection of ultra-low-power however, different studies draw different conclusion as shown
components, voltage scaling, reasonable working modes con- above description. This is maybe because placing sensors on
figuring, etc. different body segments affects particular posture recogni-
It is clear that various low-power technologies can be tion, which further reduces the performance of fall detection
used to optimize the energy consumption of fall detec- algorithm. Others focus on the influence of the sensor num-
tion system, however, most of them only use a simple ber to gain a high performance. However, different number
energy-efficient method, such as low-power hardware selec- of sensors bring out information redundancy issue. Further
tion, auto-tuning of wireless communication module. Hybrid system modeling on the body segments can be researched to
energy-efficient schemes are future trends, which combine solve above issue.
multiple low-power technologies together to optimize the
overall power consumption of the system. VI. DISCUSSION
A global study for fall detection and prevention has been
V. CLASSIFICATION ON SENSOR PLACEMENTS FOR FALL conducted by summarizing all fall-related technologies using
DETECTION variety of sensors and analytical solutions. However, no mat-
The number and placements of the wearable sensors, such as ter what types of sensors or analytical solutions have been
accelerometers or/and gyroscopes on the body have different used, there are still some limitations that need to be urgently
effects on the performance of fall detection, which have analyzed and considered.
been explored by some studies. Jacob et al. [164] presented 1) Most of previous studies on fall detection or prevention
a simple fall detection approach using one accelerometer make an ideal assumption that training data for the system
and two gyroscopes, which were placed on three different are available and sufficient to construct system model. Actu-
positions along the thoracic vertebrae to find and verify the ally, falls are abnormal events that happen infrequently and
best placement location. T-4 was indicated to be slightly diversely in real-life scenario, therefore, fall data is rather
better. Suriani et al. [45] studied the optimal sensor place- scarce and is difficult to acquire. However, in previous stud-
ment for lower activities. Accelerometer was placed on hip, ies, data used to train and verify classifier model is always
thigh, and foot to collect data for experiment verification. Hip collected from simulated falls by young people in a laboratory
was proved to be the best location to detect falls. Ntanasis environment or a controlled setting, which may be different
et al. [49] also investigated the optimal sensor placement for from actual fall data of elderly. The classifiers trained with
fall detection. Sensor locations such as the head, chest, waist, those data may suffer from high false alarm or fall missing
wrist, thigh, and ankle were studied and evaluated, among rate when they are applied into practice. In some studies, few
which the waist and the thigh were two optimal locations real fall data is collected and available in their system, how-
for high accurate fall detection. Article by [165] focused on ever, on the one hand, the collected data is still insufficient to
the impact of accelerometer number and locations on the train classifiers, on the other hand, the real data is not publicly

VOLUME 7, 2019 77717

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

available, which cannot be accessed or used to compare with researches, and also summary current low-power technique
other implementations. We believe it is benefit for the future and sensor placement of fall detection system. It is a sys-
research to incorporate real data of various falls into a public tematic study refers to all the fall-related technologies. Espe-
dataset. cially, two classification schemes for fall detection and fall
2) It has been proved that fusion-based fall detection prevention systems are proposed and conducted from sen-
and prevention systems, combining with variety of sensors sor apparatus and analytical algorithm perspectives in detail.
or analytic algorithms, work better than those single-based Each sub-category is systematically studied. Specially, from
system. However, fusion-based method integrating multiple sensor apparatus perspective, RF and fusion-based systems
data sources has redundant information, which increases the are considered. Meanwhile, from analytical algorithm aspect,
computational complexity and system cost. In this case, deter- algorithm fusion-based methods including homogeneous and
mination of the number of components to reduce redun- heterogeneous-based approaches are presented and discussed
dant information and synchronization of various sensors is detailedly. Furthermore, current challenges and issues in
inevitable to improve the overall performance of the system. fall-related systems are considered and analyzed, including
3) The main objective of current fall-related researches is lacking of real-world fall data due to various reasons, cost
achieving high performance. However, there lacks of general and information redundancy of fusion-based system, lacking
evaluation framework among different methods. Most studies of general evaluation framework, as well as high false alarm
collect their own simulated activities data in special environ- issue, all of which should be solved for future research.
ment across volunteers with different characteristics, which is
impossible to be reproduced. Therefore, it is difficult to verify REFERENCES
the given evaluation results or give a fair comparison. As it [1] World Health Organization. (2015). WHO, Number of People Over 60
is discussed earlier, age, gender, height and other external Years Set to Double by 2050; Major Societal Changes Required. [Online].
Available: https://www.who.int/mediacentre/news/releases/2015/older-
factors affect collected signals, however, these influences on persons-day/en/
proposed algorithms are not presented in most studies. Gen- [2] World Health Organization. (2007). WHO Global Report on Falls Pre-
eral evaluation framework should be discussed and presented vention in Older Age. [Online]. Available: https://www.who.int/violence
for different algorithm comparison. _injury_prevention/publications/other_injury/falls_prevention.pdf?ua=1
[3] M. C. Faes, M. F. Reelick, L. W. J.-W. Banningh, M. Gier, R. A. Esselink,
4) False alarm is also considered by fall detec- and M. G. O. Rikkert, ‘‘Qualitative study on the impact of falling in
tion/prevention system, which is ignored by current studies. frail older persons and family caregivers: Foundations for an intervention
To reduce false alarms, classifiers in fall-related systems to prevent falls,’’ Aging Mental Health, vol. 14, no. 7, pp. 834–842,
Sep. 2010.
should adapt and self-learn new activities to reduce false [4] M. Kuzuya, Y. Masuda, Y. Hirakawa, M. Iwata, H. Enoki, J. Hasegawa,
alarms, as the user has different characteristics compared with S. Izawa, and A. Iguchi, ‘‘Falls of the elderly are associated with burden of
volunteers in training phase, the collected simulated data and caregivers in the community,’’ Int. J. Geriatric Psychiatry, vol. 21, no. 8,
pp. 740–745, Aug. 2006.
the real-time running data may have different motion trend [5] S. H. Zarit, K. E. Reever, and J. Bach-Peterson, ‘‘Relatives of the impaired
and strength, which increase the false alarms of the system. elderly: Correlates of feelings of burden,’’ Gerontologist, vol. 20, no. 6,
Therefore, personalized/adaptive fall detection/prevention p. 649–655, Dec. 1980.
[6] G. Brown, ‘‘An accelerometer based fall detector: Development, exper-
system that has adaptive ability is a new trend to detect or imentation, and analysis,’’ EECS/SUPERB, Univ. California, Berkeley,
predict fall. Berkeley, CA, USA, Jul. 2005, pp. 1–9.
[7] J. Y. Hwang, J. M. Kang, Y. W. Jang, and H. C. Kim, ‘‘Development
VII. CONCLUSION AND FUTURE WORK of novel algorithm and real-time monitoring ambulatory system using
bluetooth module for fall detection in the elderly,’’ in Proc. 26th Annu. Int.
Fall detection and fall prevention systems play important Conf. IEEE Eng. Med. Biol. Soc., San Francisco, CA, USA, Sep. 2004,
roles in elderly’s daily life. Currently, various sensors are pp. 2204–2207.
deployed to determine or predict falls from ADLs, while [8] Q. Li, J. A. Stankovic, M. A. Hanson, A. T. Barth, J. Lach, and G. Zhou,
‘‘Accurate, fast fall detection using gyroscopes and accelerometer-derived
sensor-fusion method is one novel trend to improve the posture information,’’ in Proc. 6th Int. Workshop Wearable Implant. Body
performance of the system, as it combines multiple data Sensor Netw., Berkeley, CA, USA, Jun. 2009, pp. 138–143.
sources from the related sensors. From analytical algo- [9] Q. Zhang, L. Ren, and W. Shi, ‘‘HONEY: A multimodality fall detection
and telecare system,’’ Telemed. E-Health, vol. 19, no. 5, pp. 415–429,
rithm perspective, threshold-based method is a classical and May 2013.
basic approach by comparing with a reference value, while [10] A. Shahzad and K. Kim, ‘‘FallDroid: An automated smart-phone-based
machine learning-based method has been widely researched fall detection system using multiple kernel learning,’’ IEEE Trans. Ind.
Informat., vol. 15, no. 1, pp. 35–44, Jan. 2019.
to increase accuracy of the system. However, combination of [11] T. de Quadros, A. E. Lazzaretti, and F. K. Schneider, ‘‘A movement
threshold or/and machine learning-based method has sprung decomposition and machine learning-based fall detection system using
up to improve the performance of the algorithm. wrist wearable device,’’ IEEE Sensors. J., vol. 18, no. 12, pp. 5082–5089,
Jun. 2018.
In this presented paper, we conduct a comprehensive
[12] E. Cippitelli, F. Fioranelli, E. Gambi, and S. Spinsante, ‘‘Radar and RGB-
review among the latest studies on all fall-related technolo- depth sensors for fall detection: A review,’’ IEEE Sensors J., vol. 17,
gies, which is a four-layers classification. The top layer no. 12, pp. 3585–3604, Jun. 2017. doi: 10.1109/JSEN.2017.2697077.
contains four classes, including current literature reviews on [13] M. A. Habib, M. S. Mohktar, S. B. Kamaruzzaman, K. S. Lim,
T. M. Pin, and F. Ibrahim, ‘‘Smartphone-based solutions for fall detection
fall detection or/and prevention, comprehensive classifica- and prevention: Challenges and open issues,’’ Sensors, vol. 14, no. 4,
tion schemes for current fall detection and fall prevention pp. 7181–7208, 2014.

77718 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

[14] Z. Zhang, C. Conly, and V. Athitsos, ‘‘A survey on vision-based fall [36] Y. S. Delahoz and M. A. Labrador, ‘‘Survey on fall detection and fall
detection,’’ in Proc. 8th ACM Int. Conf. PErvasive Technol. Related prevention using wearable and external sensors,’’ Sensors, vol. 14, no. 10,
Assistive Environ., New York, NY, USA, Jul. 2015, p. 46. pp. 19806–19842, 2014.
[15] G. Koshmak, A. Loutfi,, and M. Linden, ‘‘Challenges and issues in [37] H. Hawley-Hague, E. Boulton, A. Hall, K. Pfeiffer, and C. Todd, ‘‘Older
multisensor fusion approach for fall detection: Review paper,’’ J. Sensors, adults’ perceptions of technologies aimed at falls prevention, detection
vol. 2016, Aug. 2015, Art. no. 6931789. doi: 10.1155/2016/6931789. or monitoring: A systematic review,’’ Int. J. Med. Inform., vol. 83, no. 6,
[16] M. Mubashir, L. Shao, and L. Seed, ‘‘A survey on fall detection: Principles pp. 416–426, Apr. 2014. doi: 10.1016/j.ijmedinf.2014.03.002.
and approaches,’’ Neurocomputing, vol. 100, pp. 144–152, Jan. 2013. [38] A. Shahzad, S. Ko, S. Lee, J.-A. Lee, and K. Kim, ‘‘Quantitative assess-
[17] F. Bagalá, C. Becker, A. Cappello, L. Chiari, K. Aminian, J. M. Hausdorff, ment of balance impairment for fall-risk estimation using wearable tri-
W. Zijlstra, and J. Klenk, ‘‘Evaluation of accelerometer-based fall detec- axial accelerometer,’’ IEEE Sensors J., vol. 17, no. 20, pp. 6743–6751,
tion algorithms on real-world falls,’’ PLoS ONE, vol. 7, no. 5, May 2012, Oct. 2017.
Art. no. e37062. [39] D. Yacchirema, J. S. de Puga, C. Palau, and M. Esteve, ‘‘Fall detection
[18] A. Bourke, P. Van de Ven, M. Gamble, R. F. O’Connor, K. Murphy, system for elderly people using IoT and big data,’’ Procedia Comput. Sci.,
E. Bogan, E. Mcquade, P. Finucane, G. Olaighin, and J. Nelson, ‘‘Evalu- vol. 130, pp. 603–610, Jan. 2018.
ation of waist-mounted tri-axial accelerometer based fall-detection algo- [40] N. Otanasap, ‘‘Pre-impact fall detection based on wearable device
rithms during scripted and continuous unscripted activities,’’ J. Biomech, using dynamic threshold model,’’ in Proc. 17th Int. Conf. Parallel Dis-
vol. 43, no. 15, pp. 3051–3057, Nov. 2010. trib. Comput., Appl. Technol. (PDCAT), Guangzhou, China, Dec. 2016,
[19] G. A. Koshmak, M. Linden, and A. Loutfi, ‘‘Evaluation of the android- pp. 362–365.
based fall detection system with physiological data monitoring,’’ in Proc. [41] N. S. Suriani, F. A. N. Rashid, and N. Y. Yunos, ‘‘Optimal accelerometer
35th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Osaka, Japan, placement for fall detection of rehabilitation patients,’’ J. Telecommun.,
Jul. 2013, pp. 1164–1168. Electron. Comput. Eng., vol. 10, nos. 2–5, pp. 25–29, Jul. 2018.
[20] C. Wang, W. Lu, M. R. Narayanan, S. J. Redmond, and N. Lovell, [42] D. Martellli, F. Artoni, V. Monaco, A. M. Sabatini, and S. Micera, ‘‘Pre-
‘‘Low-power technologies for wearable telecare and telehealth systems: impact fall detection: Optimal sensor positioning based on a machine
A review,’’ Biomed. Eng. Lett., vol. 5, no. 1, pp. 1–9, Mar. 2015. learning paradigm,’’ PLoS ONE, vol. 9, no. 3, 2014, Art. no. e92037.
[21] O. Aziz, M. Musngi, E. J. Park, G. Mori, and S. N. Robinovitch, ‘‘A com- doi: 10.1371/journal.pone.0092037.
parison of accuracy of fall detection algorithms (threshold-based vs. [43] I. P. E. S. Putra, J. Brusey, E. Gaura, and R. Vesilo, ‘‘An event-triggered
machine learning) using waist-mounted tri-axial accelerometer signals machine learning approach for accelerometer-based fall detection,’’ Sen-
from a comprehensive set of falls and non-fall trials,’’ Med. Biol. Eng. sors, vol. 18, no. 1, p. 20, Dec. 2017.
Comput., vol. 55, no. 1, pp. 45–55, Apr. 2016. doi: 10.1007/s11517-016-
[44] H. Similä, J. Mäntyjärvi, J. Merilahti, M. Lindholm, and M. Ermes,
1504-y.
‘‘Accelerometry-based berg balance scale score estimation,’’ IEEE J.
[22] J. T. Perry, S. Kellog, S. M. Vaidya, J.-H. Youn, H. Ali, and H. Sharif,
Biomed. Health Inform., vol. 18, no. 4, pp. 1114–1121, Jul. 2014.
‘‘Survey and evaluation of real-time fall detection approaches,’’ in Proc.
[45] P. Kostopoulos, T. Nunes, K. Salvi, M. Deriaz, and J. Torrent, ‘‘Increased
6th Int. Symp. High Capacity Opt. Netw. Enabling Technol. (HONET),
fall detection accuracy in an accelerometer-based algorithm consider-
Alexandria, Egypt, Dec. 2009, pp. 158–164.
ing residual movement,’’ in Proc. ICPRAM, Lisbon, Portugal, vol. 2,
[23] E. Casilari, R. Luque, and M. Morón, ‘‘Analysis of android device-based
Jan. 2015, pp. 30–36.
solutions for fall detection,’’ Sensors, vol. 15, no. 8, pp. 17827–17894,
[46] Y. Liu, S. J. Redmond, T. Shany, J. Woolgar, M. R. Narayanan, S. R. Lord,
Aug. 2015.
and N. H. Lovell, ‘‘Validation of an accelerometer-based fall prediction
[24] L. Ren, W. Shi, Z. Yu, and Z. Liu, ‘‘Real-time energy-efficient fall
model,’’ in Proc. 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.,
detection based on SSR energy efficiency strategy,’’ Int. J. Sensors Netw.,
Chicago, IL, USA, Aug. 2014, pp. 4531–4534.
vol. 20, no. 4, pp. 243–251, 2016.
[25] R. W. Broadley, J. Klenk, S. B. Thies, L. P. J. Kenney, and M. H. Granat, [47] A. Sucerquia, J. D. López, and J. F. Vargas-Bonilla, ‘‘Real-life/real-time
‘‘Methods for the real-world evaluation of fall detection technology: elderly fall detection with a triaxial accelerometer,’’ Sensors, vol. 18,
A scoping review,’’ Sensors, vol. 18, no. 7, p. 2060, Jun. 2018. no. 4, p. 1101, Apr. 2018. doi: 10.3390/s18041101.
[26] S. S. Khan and J. Hoey, ‘‘Review of fall detection techniques: A data [48] C. Medrano, R. Igual, I. Garcia-Magariño, I. Plaza, and G. Azuara, ‘‘Com-
availability perspective,’’ Med. Eng. Phys., vol. 39, pp. 12–22, Jan. 2017. bining novelty detectors to improve accelerometer-based fall detection,’’
doi: 10.1016/j.medengphy.2016.10.014. Med. Biol. Eng. Comput., vol. 55, no. 10, pp. 1849–1858, Mar. 2017.
[27] O. Mohamed, H.-J. Choi, and Y. Iraqi, ‘‘Fall detection systems for elderly doi: 10.1007/s11517-017-1632-z.
care: A survey,’’ in Proc. 6th Int. Conf. New Technol., Mobility Secur. [49] P. Perego, G. Andreoni, and G. Rizzo, Eds., ‘‘Wireless mobile communi-
(NTMS), Dubai, United Arab Emirates, Mar. 2014, pp. 1–4. cation and healthcare,’’ in MobiHealth (Lecture Notes of the Institute for
[28] S. Chaudhuri, H. Thompson, and G. Demiris, ‘‘Fall detection devices Computer Sciences, Social Informatics and Telecommunications Engi-
and their use with older adults: A systematic review,’’ J. Geriatric Phys. neering), vol. 192. Cham, Switzerland: Springer, 2016.
Therapy, vol. 37, no. 4, pp. 178–196, Oct./Dec. 2014. [50] A. Sucerquia, J. D. López, and F. Vargas, ‘‘Two-threshold energy based
[29] N. Lapierre, N. Neubauer, A. Miguel-Cruz, A. R. Rincon, L. Liu, and fall detection using a triaxial accelerometer,’’ in Proc. 38th Annu. Int.
J. Rousseau, ‘‘The state of knowledge on technologies and their use Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Orlando, FL, USA, Aug. 2016,
for fall detection: A scoping review,’’ Int. J. Med. Inform., vol. 111, pp. 3101–3104.
pp. 58–71, Mar. 2018. [51] S. Abdelhedi, R. Bourguiba, J. Mouine, and M. Baklouti, ‘‘Development
[30] P. Vallabh and R. Malekian, ‘‘Fall detection monitoring systems: A com- of a two-threshold-based fall detection algorithm for elderly health moni-
prehensive review,’’ J. Ambient Intell. Humanized Comput., vol. 9, no. 6, toring,’’ IEEE 10th Int. Conf. Res. Challenges Inf. Sci. (RCIS), Grenoble,
pp. 1809–1833, Nov. 2018. France, Jun. 2016, pp. 1–5.
[31] X. Hu and X. Qu, ‘‘Pre-impact fall detection,’’ BioMed. Eng. OnLine, [52] C. Medrano, R. Igual, I. Plaza, and M. Castro, ‘‘Detecting falls as
vol. 15, no. 61, Jun. 2016. novelties in acceleration patterns acquired with smartphones,’’ PLoS
[32] N. M. Kosse, K. Brands, J. M. Bauer, T. Hortobagyi, and C. J. Lamoth, ONE, vol. 9, no. 4, Apr. 2014, Art. no. e94811. doi: 10.1371/journal.
‘‘Sensor technologies aiming at fall prevention in institutionalized old pone.0094811.
adults: A synthesis of current knowledge,’’ Int. J. Med. Inform., vol. 82, [53] M. Á. Á. de la Concepcion, L. M. S. Morillo, J. A. Álvarez-García, and
no. 9, pp. 743–752, Sep. 2013. L. González-Abril, ‘‘Mobile activity recognition and fall detection system
[33] R. Sun and J. J. Sosnoff, ‘‘Novel sensing technology in fall risk assess- for elderly people using ameva algorithm,’’ Pervasive Mobile Comput.,
ment in older adults: A systematic review,’’ BMC Geriatrics, vol. 18, vol. 34, pp. 3–13, Jan. 2017. doi: 10.1016/j.pmcj.2016.05.002.
p. 14, Jan. 2018. doi: 10.1186/S12877-018-0706-6. [54] J. He, C. Hu, and X. Wang, ‘‘A smart device enabled system for
[34] A. Oladele, S. Antony, and S. June, ‘‘Use of information technology for autonomous fall detection and alert,’’ Int. J. Distrib. Sensor Netw.,
falls detection and prevention in the elderly,’’ Ageing Int., vol. 40, no. 3, vol. 2016, Feb. 2016, Art. no. 2308183. doi: 10.1155/2016/2308183.
pp. 277–299, Sep. 2014. doi: 10.1007/s12126-014-9204-0. [55] R. M. Gibson, A. Amira, N. Ramzan, P. Casaseca-de-la-Higuera, and
[35] K. Chaccour, R. Darazi, A. H. El Hassani, and E. Andrès, ‘‘From fall Z. Pervez, ‘‘Multiple comparator classifier framework for accelerometer-
detection to fall prevention: A generic classification of fall-related sys- based fall detection and diagnostic,’’ Appl. Soft. Comput., vol. 39,
tems,’’ IEEE Sensors J., vol. 17, no. 3, pp. 812–821, Feb. 2017. pp. 94–103, Feb. 2016.

VOLUME 7, 2019 77719

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

[56] L. Palmerini, F. Bagalá, A. Zanetti, J. Klenk, C. Becker, and A. Cappello, [77] A. Leone, G. Rescio, A. Caroppo, and P. Siciliano, ‘‘An EMG-based
‘‘A wavelet-based approach to fall detection,’’ Sensors, vol. 15, no. 5, system for pre-impact fall detection,’’ in Proc. IEEE Sensors, Busan,
pp. 11575–11586, May 2015. South Korea, Nov. 2015, pp. 1–4.
[57] D. Lim, C. Park, N. H. Kim, S.-H. Kim, and Y. S. Yu, ‘‘Fall-detection [78] A. Leone, G. Rescio, A. Caroppo, and P. Siciliano, ‘‘A wearable
algorithm using 3-axis acceleration: Combination with simple thresh- EMG-based system pre-fall detector,’’ Procedia Eng., vol. 120,
old and hidden Markov model,’’ J. Appl. Math., vol. 2014, Sep. 2014, pp. 455–458, Jan. 2015.
Art. no. 896030. doi: 10.1155/2014/896030. [79] V. F. Annese, M. Crepaldi, D. Demarchi, and D. De Venuto, ‘‘A digital
[58] N. Pannurat, S. Thiemjarus, and E. Nantajeewarawat, ‘‘A hybrid temporal processor architecture for combined EEG/EMG falling risk prediction,’’
reasoning framework for fall monitoring,’’ IEEE Sensors J., vol. 17, no. 6, in Proc. Design, Automat. Test Eur. Conf. Exhib. (DATE), Dresden, Ger-
pp. 1749–1759, Mar. 2017. many, Mar. 2016, pp. 714–719.
[59] M. Hemmatpour, M. Karimshoushtari, R. Ferrero, B. Montrucchio, [80] J. Sun, Z. Wang, L. Chen, B. Wang, C. Ji, and S. Tao, ‘‘A plantar
M. Rebaudengo, and C. Novara, ‘‘Polynomial classification model for inclinometer based approach to fall detection in open environments,’’ in
real-time fall prediction system,’’ in Proc. IEEE 41st Annu. Comput. Proc. Emerg. Trends Adv. Technol. Comput. Intell., Extended Sel. Results
Softw. Appl. Conf. (COMPSAC), Turin, Italy, Jul. 2017, pp. 973–978. Sci. Inf. Conf., vol. 647, Jun. 2016, pp. 1–13.
[60] R.-K. Shen, C.-Y. Yang, V. R. L. Shen, and W.-C. Chen, ‘‘A novel fall [81] J. Sun, Z. Wang, B. Pei, S, Tao, L. Chen, ‘‘Fall detection using plantar
prediction system on smartphones,’’ IEEE Sensors J., vol. 17, no. 6, inclinometer sensor,’’ in Proc. IEEE 12th Intl Conf. Ubiquitous Intell.
pp. 1865–1871, Mar. 2017. Comput., IEEE 12th Int. Conf. Autonomic Trusted Comput., IEEE 15th
[61] L. Mao, D. Liang, Y. Ning, Y. Ma, X. Gao, and G. Zhao, ‘‘Pre-impact Int. Conf, Scalable Comput., Commun. Associated Workshops (UIC-ATC-
and impact detection of falls using built-in tri-accelerometer of smart- ScalCom), Beijing, China, Aug. 2015, pp. 1692–1697.
phone,’’ in Health Information Science. Shenzhen, China: Springer, 2014, [82] D. Droghini, E. Principi, S. Squartini, P. Olivetti, and F. Piazza, ‘‘Human
pp. 167–174. fall detection by using an innovative floor acoustic sensor,’’ Smart Innov.,
[62] J. Fontecha, F. J. Navarro, R. Hervás, and J. Bravo, ‘‘Elderly frailty detec- vol. 69, pp. 97–107, Aug. 2017.
tion by using accelerometer-enabled smartphones and clinical informa- [83] A. Irtaza, S. M. Adnan, S. Aziz, A. Javed, M. O. Ullah, and
tion records,’’ Pers. Ubiquitous Comput., vol. 17, no. 6, pp. 1073–1083, M. T. Mahmood, ‘‘A framework for fall detection of elderly people by
Aug. 2013. analyzing environmental sounds through acoustic local ternary patterns,’’
[63] F. Riva, M. J. P. Toebes, M. Pijnappels, R. Stagni, and J. H. van in Proc. IEEE Int. Conf. Syst., Man, Cybern. (SMC), Banff, AB, Canada,
Dieen, ‘‘Estimating fall risk with inertial sensors using gait stability Oct. 2017, pp. 1558–1563.
measures that do not require step detection,’’ Gait Posture, vol. 38, no. 2, [84] S. M. Adnan, A. Irtaza, S. Aziz, M. O. Ullah, A. Javed, and
pp. 170–174, Jun. 2013. M. T. Mahmood, ‘‘Fall detection through acoustic local ternary patterns,’’
[64] A. K. Bourke and G. M. Lyons, ‘‘A threshold-based fall-detection algo- Appl. Acoust., vol. 140, pp. 296–300, Nov. 2018.
rithm using a bi-axial gyroscope sensor,’’ Med. Eng. Phys., vol. 30, no. 1, [85] D. Droghini, D. Ferretti, E. Principi, S. Squartini, and F. Piazza, ‘‘A com-
pp. 84–90, Jan. 2008. bined one-class SVM and Template-matching approach for user-aided
human fall detection by means of floor acoustic features,’’ Comput. Intell.
[65] Y. Su, D. Liu, and Y. Wu, ‘‘A multi-sensor based pre-impact fall detection
Neurosci., vol. 2017, Jan. 2017, Art. no. 1512670.
system with a hierarchical classifier,’’ in Proc. 9th Int. Congr. Image
[86] M. S. Khan, M. Yu, P. Feng, L. Wang, and J. Chambers, ‘‘An unsupervised
Signal Process., BioMed. Eng. Inform. (CISP-BMEI), Datong, China,
acoustic fall detection system using source separation for sound interfer-
Oct. 2016, pp. 1727–1731.
ence suppression,’’ Signal Process., vol. 110, pp. 199–210, May 2015.
[66] O. Almeida, M. Zhang, and J.-C. Liu, ‘‘Dynamic fall detection and
[87] Y. Li, K. C. Ho, M. Popescu, and M. Skubic, ‘‘A theoretical study on the
pace measurement in walking sticks,’’ in Proc. Joint Workshop High
placement of microphone arrays for improving the localization accuracy
Confidence Med. Devices, Softw., Syst. Med. Device Plug-and-Play
of a fall,’’ in Proc. 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.,
Interoperability (HCMDSS-MDPnP), Washington, DC, USA, Jun. 2007,
Chicago, IL, USA, Aug. 2014, pp. 4523–4526.
pp. 204–206.
[88] Y. Li, K. C. Ho, and M. Popescu, ‘‘A microphone array system for
[67] P. C. Fino, C. W. Frames, and T. E. Lockhart, ‘‘Classifying step and
automatic fall detection,’’ IEEE Trans. Biomed. Eng., vol. 59, no. 2,
spin turns using wireless gyroscopes and implications for fall risk assess-
pp. 1291–1301, May 2012.
ments,’’ Sensors, vol. 15, pp. 10676–10685, May 2015.
[89] K. Chaccour, R. Darazi, A. H. el Hassans, and E. Andres, ‘‘Smart
[68] W. Lu, C. Wang, M. C. Stevens, S. J. Redmond, and N. H. Lovell, ‘‘Low- carpet using differential piezoresistive pressure sensors for elderly fall
power operation of a barometric pressure sensor for use in an automatic detection,’’ in Proc. IEEE 11th Int. Conf. Wireless Mobile Comput.,
fall detector,’’ in Proc. 38th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. Netw. Commun. (WiMob), Abu Dhabi, United Arab Emirates, Oct. 2015,
(EMBC), Orlando, FL, USA, Aug. 2016, pp. 2010–2013. pp. 225 –229.
[69] K. Chaccour, H. Al Assaad, A. H. El Hassani, R. Darazi, and E. Andres, [90] J. M. A. Morgado and A. König, ‘‘Low-power concept and prototype of
‘‘Sway analysis and fall prediction method based on spatio-temporal distributed resistive pressure sensor array for smart floor and surfaces in
sliding window technique,’’ in Proc. IEEE 18th Int. Conf. e-Health Netw., intelligent environments,’’ in Proc. Int. Multi-Conf. Syst., Signals Devices,
Appl. Services (Healthcom), Munich, Germany, Sep. 2016, pp. 1–6. Chemnitz, Germany, Mar. 2012, pp. 1–6.
[70] J. Light, S. Cha, and M. Chowdhury, ‘‘Optimizing pressure sensor array [91] F. Muheidat, L. Tawalbeh, and H. Tyrer, ‘‘Context-aware, accurate, and
data for a smart-shoe fall monitoring system,’’ in Proc. IEEE Sensors, real time fall detection system for elderly people,’’ in Proc. IEEE 12th
Busan, South Korea, Nov. 2015, pp. 1–4. Int. Conf. Semantic Comput. (ICSC), Laguna Hills, CA, USA, Feb. 2018,
[71] J. Verghese, R. Holtzer, R. B. Lipton, and C. Wang, ‘‘Quantitative gait pp. 329–333.
markers and incident fall risk in older adults,’’ J. Gerontol., Med. Sci., [92] D. Mcgrath, E. P. Doheny, L. Walsh, D. J. McKeown, C. Cunningham,
vol. 64A, no. 8, pp 896–901, Aug. 2009. L. Crosby, R. A. Kenny, N. Stergiou, B. Caulfield, and B. R. Greene,
[72] H. Han, X. Ma, and K. Oyama, ‘‘Flexible detection of fall events using ‘‘Taking balance measurement out of the laboratory and into the home:
bidirectional EMG sensor,’’ Stud. Health Technol. Inf., vol. 245, no. 2017, Discriminatory capability of novel centre of pressure measurement in
p. 1225, Jan. 2017. doi: 10.3233/978-1-61499-830-3-1225. fallers and non-fallers,’’ in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol.
[73] G. Rescio, A. Leone, and P. Siciliano, ‘‘Supervised machine learning Soc., San Diego, CA, USA, Aug./Sep. 2012.
scheme for electromyography-based pre-fall detection system,’’ Expert [93] Y. Zigel, D. Litvak, and I. Gannot, ‘‘A method for automatic fall detection
Syst. Appl., vol. 100, pp. 95–105, Jun. 2018. of elderly people using floor vibrations and sound—Proof of concept on
[74] X. Xi, M. Tang, S. M. Miran, and Z. Luo, ‘‘Evaluation of feature extrac- human mimicking doll falls,’’ IEEE Trans. Biomed. Eng., vol. 56, no. 12,
tion and recognition for activity monitoring and fall detection based on pp. 2858–2867, Dec. 2009.
wearable sEMG sensors,’’ Sensors, vol. 17, no. 6, p. 1229, May 2017. [94] H. Rimminen, J. Lindström, M. Linnavuo, and R. Sepponen, ‘‘Detection
[75] A. Leone, G. Rescio, and P. Siciliano, ‘‘Fall risk evaluation by surface of falls among the elderly by a floor sensor using the electric near
electromyography technology,’’ in Proc. Int. Conf. Eng., Technol. Innov. field,’’ IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 6, pp. 1475–1476,
(ICE/ITMC), Funchal, Portuga, Jun. 2017, pp. 1092–1095. Jun. 2010.
[76] J. Van Vaerenbergh, L. Briers, and P. Broos, ‘‘Electromyography study [95] T. Marasovic, M. Cecić, and V. Zanchi, ‘‘Analysis and interpretation of
of the anterior tibial muscle and accidental falls in the elderly,’’ Paxis, ground reaction forces in normal gait,’’ WSEAS Trans. Syst., vol. 8, no. 9,
vol. 90, no. 50, pp. 2199–2204, Dec. 2001. pp. 1105–1114, Jan. 2009.

77720 VOLUME 7, 2019

L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

[96] R. G. Cumming, C. Sherrington, S. R. Lord, J. M. Simpson, C. Vogler, [117] S. Vadivelu, S. Ganesan, O. V. R. Murthy, and A. Dhall, ‘‘Thermal imag-
I. D. Cameron, and V. Naganathan, ‘‘Cluster randomised trial of a tar- ing based elderly fall detection,’’ in Comuter Vision—ACCV, vol. 10118.
geted multifactorial intervention to prevent falls among older people in Cham, Switzerland: Springer, 2016, pp. 541–553.
hospital,’’ Brit. Med. J., vol. 336, pp. 758–760, Apr. 2008. [118] Y. Tian, G. H. Lee, H. He, C. Y. Hsu, and D. Katabi, ‘‘RF-based fall
[97] X. Fan, H. Zhang, C. Leung, and Z. Shen, ‘‘Robust unobtrusive fall monitoring using convolutional neural networks,’’ Proc. ACM Interact.,
detection using infrared array sensors,’’ in Proc. IEEE Int. Conf. Multi- Mobile, Wearable Ubiquitous Technologie, vol. 2, no. 3, p. 137, Sep. 2018.
sensor Fusion Integr. Intell. Syst. (MFI), Daegu, South Korea, Nov. 2017, doi: 10.1145/3264947.
pp. 194–199. [119] Y. Tang, Z. Peng, and C. Li, ‘‘An experimental study on the feasibility
[98] S. Nishiguchi, M. Yamada, K. Uemura, T. Matsumura, M. Takahashi, of fall prevention using a wearable k-band FMCW radar,’’ in Proc.
T. Moriguchi, and T. Aoyama, ‘‘A novel infrared laser device that mea- United States Nat. Committee URSI Nat. Radio Sci. Meeting (USNC-
sures multilateral parameters of stepping performance for assessment of URSI NRSM), Boulder, CO, USA, Jan. 2017, pp. 1–2.
fall risk in elderly individuals,’’ Aging Clin. Exp. Res., vol. 25, no. 3, [120] B. Jokanović and M. Amin, ‘‘Fall detection using deep learning in range-
pp. 311–316, Jun. 2013. Doppler radars,’’ IEEE Trans. Aerosp. Electron. Syst., vol. 54, no. 1,
[99] J. Güttler, A. Bittner, K. Langosch, T. Bock, and Y. Mitsukura, ‘‘Devel- pp. 180–189, Feb. 2018.
opment of an affordable and easy-to-install fall detection system,’’ IEEJ [121] Y. Tang, Z. Peng, L. Ran, and C. Li, ‘‘iPrevent: A novel wearable
Trans. Electr. Electron. Eng., vol. 13, no. 5, pp. 664–670, May 2018. radio frequency range detector for fall prevention,’’ in Proc. IEEE Int.
[100] A. Yazar, F. Keskin, B. U. Töreyin, and A. E. Çetin, ‘‘Fall detection using Symp. Radio-Freq. Integr. Technol. (RFIT), Taipei, Taiwan, Aug. 2016,
single-tree complex wavelet transform,’’ Pattern Recognit. Lett., vol. 34, pp. 1–3.
no. 15, pp. 1945–1952, Nov. 2013. [122] K. Shiba, T. Kaburagi, and Y. Kurihara, ‘‘Fall detection utilizing fre-
[101] M. Popescu, B. Hotrabhavananda, M. Moore, and M. Skubic, quency distribution trajectory by microwave Doppler sensor,’’ IEEE Sen-
‘‘VAMPIR—An automatic fall detection system using a vertical sors J., vol. 17, no. 22, pp. 7561–7568, Nov. 2017.
PIR sensor array,’’ in Proc. 6th Int. Conf. Pervasive Comput. Technol. [123] R. Visvanathan, D. C. Ranasinghe, R. L. S. Torres, and K. Hill, ‘‘Frame-
Healthcare (PervasiveHealth) Workshops, San Diego, CA, USA, work for preventing falls in acute hospitals using passive sensor enabled
May 2012, pp. 163–166. radio frequency identification technology,’’ in Proc. Annu. Int. Conf.
[102] K. de Miguel, A. Brunete, M. Hernando, and E. Gambao, ‘‘Home camera- IEEE Eng. Med. Biol. Soc., San Diego, CA, USA, Aug./Sep. 2012,
based fall detection system for the elderly,’’ Sensors, vol. 17, no. 12, pp. 5858–5862.
p. 2864, Dec. 2017. [124] H. Wang, D. Zhang, Y. Wang, J. Ma, Y. G. Wang, and S. Li, ‘‘RT-
[103] J. Kutchka, D. Tchuinkou, J. Mandebi, E. Nghonda, and C. Bobda, Fall: A real-time and contactless fall detection system with commodity
‘‘Automatic assessment of environmental hazards for fall prevention WiFi devices,’’ IEEE Trans. Mobile Comput., vol. 16, no. 2, p. 511–526,
using smart-cameras,’’ in Proc. IEEE 1st Int. Conf. Connected Health, Feb. 2017.
Appl., Syst. Eng. Technol. (CHASE), Washington, DC, USA, Jun. 2016, [125] Y. Wang, K. Wu, and L. M. Ni, ‘‘WiFall: Device-free fall detection
pp. 24–29. by wireless networks,’’ IEEE Trans. Mobile Comput., vol. 16, no. 2,
[104] K. Fan, P. Wang, Y. Hu, and B. Dou, ‘‘Fall detection via human pos- pp. 581–594, Feb. 2017.
ture representation and support vector machine,’’ Int. J. Distrib. Sen- [126] J. Howcroft, J. Kofman, and E. D. Lemaire, ‘‘Prospective fall-risk
sor Netw., vol. 13, no. 5, May 2017, Art. no. 1550147717707418. prediction models for older adults based on wearable sensors,’’ IEEE
doi: 10.1177/1550147717707418. Trans. Neural Syst. Rehabil. Eng., vol. 25, no. 10, pp. 1812–1820,
[105] T.-H. Li, C.-Y. Yang, and W.-P. Shih, ‘‘Fall prevention shoes using Oct. 2017.
camera-based line-laser obstacle detection system,’’ J. Healthcare Eng., [127] Y. Wu, Y. Su, Y. Hu, N. Yu, and R. Feng, ‘‘A multi-sensor fall detection
vol. 2017, Apr. 2017, Art. no. 8264071. system based on multivariate statistical process analysis,’’ J. Med. Biol.
[106] A. Manzi, F. Cavallo, and P. Dario, ‘‘A neural network approach to Eng., vol. 39, no. 3, pp. 336–351, May 2018. doi: 10.1007/s40846-018-
human posture classification and fall detection using RGB-D camera,’’ 0404-z.
in Ambient Assisted Living, vol. 426. Cham, Switzerland: Springer, 2016, [128] A. Leone, G. Rescio, A. Caroppo, and P. Siciliano, ‘‘Wireless electromyo-
pp. 127–139. graphy technology for fall risk evaluation,’’ Convegno Nazionale Sensori,
[107] K. Makantasis, E. Protopapadakis, A. Doulamis, N. Doulamis, and vol. 431, pp. 322–329, 2017. doi: 10.1007/978-3-319-55077-0_41.
N. Matsatsinis, ‘‘3D measures exploitation for a monocular semi- [129] A. Ejupi, C. Galang, O. Azizi, E. J. Park, and S. Robinovitch, ‘‘Accu-
supervised fall detection system,’’ Multimedia Tools Appl., vol. 75, no. 22, racy of a wavelet-based fall detection approach using an accelerome-
pp. 15017–15049, Nov. 2015. ter and a barometric pressure sensor,’’ in Proc. 39th Annu. Int. Conf.
[108] K. Bradley, ‘‘Remote video monitoring: A novel approach in fall pre- IEEE Eng. Med. Biol. Soc. (EMBC), Seogwipo, South Korea, Jul. 2017,
vention,’’ J. Continuing Educ. Nursing, vol. 47, no. 11, pp. 484–486, pp. 2150–2153.
Nov. 2016. [130] M. Hemmatpour, R. Ferrero, B. Montrucchio, and M. Rebaudengo,
[109] F. Zhao, Z. Cao, Y. Xiao, J. Mao, and J. Yuan, ‘‘Real-time detection of fall ‘‘A neural network model based on co-occurrence matrix for fall predic-
from bed using a single depth camera,’’ IEEE Trans. Autom. Sci. Eng., to tion,’’ in Proc. Int. Conf. Wireless Mobile Commun., vol. 192, Jun. 2017,
be published. doi: 10.1109/TASE.2018.2861382. pp. 241–248.
[110] M. Li, G. Xu, B. He, X. Ma, and J. Xie, ‘‘Pre-impact fall detection [131] C. Lu, J. Huang, Z. Lan, and Q. Wang, ‘‘Bed exiting monitoring
based on a modified zero moment point criterion using data from kinect system with fall detection for the elderly living alone,’’ in Proc. Int.
sensors,’’ IEEE Sensors J., vol. 18, no. 13, pp. 5522–5531, Jul. 2018. Conf. Adv. Robot. Mechatronics (ICARM), Macau, China, Aug. 2016,
[111] X. Kong, L. Meng, and H. Tomiyama, ‘‘Fall detection for elderly persons pp. 59–64.
using a depth camera,’’ Proc. Int. Conf. Adv. Mechtron. Syst. (ICAMechS), [132] A. J. A. Majumder, I. Zerin, S. I. Ahamed, and R. O. Smith,
Xiamen, China, Dec. 2017, pp. 269–273. ‘‘A multi-sensor approach for fall risk prediction and prevention in
[112] X. Tao and Z. Yun, ‘‘Fall prevention based on biomechanics equilibrium elderly,’’ ACM SIGAPP Appl. Comput. Rev., vol. 14, no. 1, pp. 41–52,
using kinect,’’ Int. J. Distrib. Sensors Netw., vol. 13, no. 4, Apr. 2017, Mar. 2014.
Art. no. 1550147717703257. doi: 10.1177/1550147717703257. [133] J. He, S. Bai, and X. Wang, ‘‘An unobtrusive fall detection and alerting
[113] E. Akagündüz, M. Aslan, A. Şengur, H. Wang, and M. C. Ince, ‘‘Silhou- system based on Kalman filter and Bayes network classifier,’’ Sensors,
ette orientation volumes for efficient fall detection in depth videos,’’ IEEE vol. 17, no. 6, p. 1393, Jun. 2017.
J. Biomed. Health Inform., vol. 21, no. 3, pp. 756–763, May 2017. [134] A. K. Thella, V. K. Suryadevara, M. Rizkalla, and G. Hossain, ‘‘Smart unit
[114] A. Dubois and F. Charpillet, ‘‘A gait analysis method based on a depth care for pre fall detection and prevention,’’ in Proc. IEEE Nat. Aerosp.
camera for fall prevention,’’ in Proc. 36th Annu. Int. Conf. IEEE Eng. Electron. Conf. (NAECON), Ohio Innov. Summit (OIS), Dayton, OH,
Med. Biol. Soc., Chicago, IL, USA, Aug. 2014, pp. 4515–4518. USA, Jul. 2016, pp. 220–224.
[115] J. Rafferty, J. Synnott, C. Nugent, G. Morrison, and E. Tamburini, ‘‘Fall [135] M. Kepski and B. Kwolek, ‘‘Event-driven system for fall detection using
detection through thermal vision sensing,’’ in Ubiquitous Computing and body-worn accelerometer and depth sensor,’’ IET Comput. Vis., vol. 12,
Ambient Intelligence. Cham, Switzerland: Springer, 2016, pp. 84–90. no. 1, pp. 48–58, Feb. 2018.
[116] K.-S. Song, Y.-H. Nho, and D.-S. Kwon, ‘‘Histogram based fall pre- [136] R. Ramezani, Y. Xiao, and A. Naeim, ‘‘Sensing-Fi: Wi-Fi CSI and
diction of patients using a thermal imagery camera,’’ in Proc. 14th Int. accelerometer fusion system for fall detection,’’ in Proc. IEEE EMBS Int.
Conf. Ubiquitous Robots Ambient Intell. (URAI), Jeju, South Korea, Conf. Biomed. Health Inform. (BHI), Las Vegas, NV, USA, Mar. 2018,
Jun./Jul. 2017, pp. 161–164. pp. 402–405.

VOLUME 7, 2019 77721

 L. Ren, Y. Peng: Research of Fall Detection and Fall Prevention Technologies: A Systematic Review

[137] A. Ejupi, Y. J. Gschwind, T. Valenzuela, S. R. Lord, and [155] W.-C. Cheng and D.-M. Jhan, ‘‘Triaxial accelerometer-based fall detec-
K. Delbaere, ‘‘A kinect and inertial sensor-based system for the tion method using a self-constructing cascade-adaboost-SVM classifier,’’
self-assessment of fall risk: A home-based study in older people,’’ IEEE J. Biomed. Health Inform., vol. 17, no. 2, pp. 411–417, Mar. 2013.
Hum.-Comput. Interact., vol. 31, nos. 3–4, pp. 261–293, 2016. [156] S. B. Khojasteh, J. R. Villar, C. Chira, V. M. Gonzàlez, and E. de la Cal,
doi: 10.1080/07370024.2015.1085309. ‘‘Improving fall detection using an on-wrist wearable accelerometer,’’
[138] H. Li, A. Shrestha, F. Fioranelli, J. L. Kernec, H. Heidari, M. Pepa, Sensors, vol. 18, no. 5, p. 1350, Apr. 2018.
E. Cippitelli, E. Gambi, and S. Spinsante, ‘‘Multisensor data fusion for [157] X. Li, L. Nie, H. Xu, and X. Wang, ‘‘Collaborative fall detection using
human activities classification and fall detection,’’ in Proc. IEEE Sensors, smart phone and kinect,’’ Mobile Netw. Appl., vol. 23, no. 4, pp. 775–788,
Glasgow, U.K., Nov. 2017, pp. 1–3. Aug. 2018.
[139] M. Rantz, M. Skubic, C. Abbott, C. Galambos, M. Popescu, J. Keller, [158] C. Wang, M. R. Narayanan, S. R. Lord, S. J. Redmond, and N. H. Lovell,
E. Stone, J. Back, S. J. Miller, and G. F. Petroski, ‘‘Automated in-home ‘‘A low-power fall detection algorithm based on triaxial acceleration and
fall risk assessment and detection sensor system for elders,’’ Gerontolo- barometric pressure,’’ in Proc. 36th Annu. Int. Conf. IEEE Eng. Med. Biol.
gist, vol. 55, pp. S78–S87, Jun. 2015. doi: 10.1093/geront/gnv044. Soc., Chicago, IL, USA, Aug. 2014, pp. 570–573.
[140] D. Razum, G. Seketa, J. Vugrin, and I. Lackovic, ‘‘Optimal threshold [159] M. G. de al Solaz, A. Bourke, R. Conway, J. Nelson, and G. ÓLaighin,
selection for threshold-based fall detection algorithms with multiple fea- ‘‘Real-time low-energy fall detection algorithm with a programmable
tures,’’ in Proc. 41st Int. Conv. Inf. Commun. Technol., Electron. Micro- truncated MAC,’’ in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol.,
electron. (MIPRO), Opatija, Croatia, May 2018, pp. 1513–1516. Aug./Sep. 2010, pp. 2423–2426.
[141] M. Hemmatpour, R. Ferrero, F. Gandino, B. Montrucchio, and [160] J. Yuan, K. K. Tan, T. H. Lee, and G. C. H. Koh, ‘‘Power-efficient
M. Rebaudengo, ‘‘Nonlinear predictive threshold model for real-time interrupt-driven algorithms for fall detection and classification of activ-
abnormal gait detection,’’ J. Healthcare Eng., vol. 2018, Jun. 2018, ities of daily living,’’ IEEE Sensors J., vol. 15, no. 3, pp. 1377–1387,
Art. no. 4750104. doi: 10.1155/2018/4750104. Mar. 2015.
[142] N. P. Pham, H. V. Dao, H. N. Phung, H. Van Ta, N. H. Nguyen, and [161] T. N. Gia, V. Sarker, I. Tcarenko, A. M. Rahmani, T. Westerlund,
T. T. Hoang, ‘‘Classification different types of fall for reducing false P. Liljeberg, and H. Tenhunen, ‘‘Energy efficient wearable sensor node for
alarm using single accelerometer,’’ in Proc. IEEE 7th Int. Conf. Commun. IoT-based fall detection systems,’’ Elsevier Microprocessors Microsyst.,
Electron. (ICCE), Hue, Vietnam, Jul. 2018, pp. 316–321. vol. 56, pp. 34–36, Feb. 2018. doi: 10.1016/j.micpro.2017.10.014.
[143] T. Sivaranjani, L. DhiviyaLakshmi, R. Yogaaravinth, J. Srivishnu, [162] T. N. Gia, I. Tcarenko, V. K. Sarker, A. M. Rahmani, T. Westerlund,
M. M. S. Karthick, and A. Praveenkumar, ‘‘Fall assessment and its injury P. Liljeberg, and H. Tenhunen, ‘‘IoT-based fall detection system with
prevention using a wearable airbag technology,’’ in Proc. IEEE Int. energy efficient sensor nodes,’’ in Proc. IEEE Nordic Circuits Syst. Conf.
Conf. Power, Control, Signals Instrum. Eng. (ICPCSI), Chennai, India, (NORCAS), Copenhagen, Denmark, Nov. 2016, pp. 1–6.
Sep. 2017, pp. 2539–2541. [163] C. Wang, W. Lu, M. R. Narayanan, D. C. W. Chang, S. R. Lord,
[144] L. Ren and W. Shi, ‘‘Chameleon: Personalised and adaptive fall detection S. J. Redmond, and N. H. Lovell, ‘‘Low-power fall detector using triaxial
of elderly people in home-based environments,’’ Int. J. Sens. Netw., accelerometry and barometric pressure sensing,’’ IEEE Trans. Ind. Infor-
vol. 20, no. 3, pp. 163–176, 2016. mat., vol. 12, no. 6, pp. 2302–2311, Dec. 2016.
[145] Y. Cao, Y. Yang, and W. Liu, ‘‘E-FallD: A fall detection system using [164] J. Jacob, T. Nguyen, D. Y. C. Lie, S. Zupancic, J. Bishara, A. Dentino, and
android-based smartphone,’’ in Proc. 9th Int. Conf. Fuzzy Syst. Knowl. R. E. Banister, ‘‘A fall detection study on the sensors placement location
Discovery, Sichuan, China, May 2012. and a rule-based multi-thresholds algorithm using both accelerometer and
[146] S. Yu, H. Chen, and R. A. Brown, ‘‘Hidden Markov model-based fall gyroscopes,’’ in Proc. IEEE Int. Conf. Fuzzy Syst. (FUZZ-IEEE), Taipei,
detection with motion sensor orientation calibration: A case for real- Taiwan, Jun. 2017, pp. 666–671.
life home monitoring,’’ IEEE J. Biomed. Health Inform., vol. 22, no. 6, [165] H. Gjoreski, M. Lustrek, and M. Cams, ‘‘Accelerometer placement for
pp. 1847–1853, Nov. 2018. posture recognition and fall detection,’’ in Proc. 7th Int. Conf. Intell.
[147] X. Zhou, L.-C. Qian, P.-J. You, Z.-G. Ding, and Y.-Q. Han, ‘‘Fall detection Environ., Jul. 2011, pp. 47–54.
using convolutional neural network with multi-sensor fusion,’’ in Proc.
IEEE Int. Conf. Multimedia Expo Workshops (ICMEW), San Diego, CA,
USA, Jul. 2018, pp. 1–5.
[148] T. Zhen, L. Mao, J. Wang, and Q. Gao, ‘‘Wearable preimpact fall detector LINGMEI REN received the Ph.D. degree from
using SVM,’’ in Proc. 10th Int. Conf. Sens. Technol. (ICST), Nanjing, the Department of Electronic and Information
China, Nov. 2016, pp. 1–6. Engineering, Tongji University, in 2016. She was
[149] W. Min, H. Cui, H. Rao, Z. Li, and L. Yao, ‘‘Detection of human falls
a Visiting Scholar with Wayne State University,
on furniture using scene analysis based on deep learning and activity
characteristics,’’ IEEE Access, vol. 6, pp. 9324–9335, 2018.
in 2012. She is currently with the Department of
[150] L. Steffan, L. Kaul, and T. Asfour, ‘‘Online stability estimation based on Computer Science, Shandong University of Sci-
inertial sensor data for human and humanoid fall prevention,’’ in Proc. ence and Technology, Qingdao, China. Her main
IEEE-RAS 17th Int. Conf. Humanoid Robot. (Humanoids), Birmingham, research interests include fall detection, human
U.K., Nov. 2017, pp. 171–177. behavior recognition, wireless health, and edge
[151] W. Min, L. Yao, Z. Lin, and L. Liu, ‘‘Support vector machine approach to computing.
fall recognition based on simplified expression of human skeleton action
and fast detection of start key frame using torso angle,’’ IET Comput. Vis.,
vol. 12, no. 8, pp. 1133–1140, Jul. 2018. doi: 10.1049/iet-cvi.2018.5324.
[152] O. Aziz, C. M. Russell, E. J. Park, and S. N. Robinovitch, ‘‘The effect
YANJUN PENG received the Ph.D. degree,
of window size and lead time on pre-impact fall detection accuracy using
in 2004. He joined the Department of Computer
support vector machine analysis of waist mounted inertial sensor data,’’
in Proc. 36th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., Aug. 2014, Science, Shandong University of Science and
pp. 30–33. Technology, Qingdao, China, in 1996, where he
[153] X. Hu and X. Qu, ‘‘An individual-specific fall detection model based was promoted to Professor, in 2010. His main
on the statistical process control chart,’’ Saf. Sci., vol. 64, pp. 13–21, research interests include medicine visualization,
Apr. 2014. virtual reality, and image processing.
[154] A. Poonsri and W. Chiracharit, ‘‘Improvement of fall detection using
consecutive-frame voting,’’ in Proc. Int. Workshop Adv. Image Technol.
(IWAIT), Chiang Mai, Thailand, Jan. 2018, pp. 1–4.

77722 VOLUME 7, 2019