Content uploaded by Reza Malekian
Author content
All content in this area was uploaded by Reza Malekian on Dec 28, 2017
Content may be subject to copyright.
Noname manuscript No.
(will be inserted by the editor)
Fall detection monitoring systems: A comprehensive review
Pranesh Vallabh ·Reza Malekian
Received: date / Accepted: date
Abstract The increase in elderly population especially
in the developed countries and the number of elderly
people living alone can result in increased healthcare
costs which can cause a huge burden on the society.
With fall being one of the biggest risk among the elderly
population resulting in serious injuries, if not treated
quickly. The advancements in technology, over the years,
resulted in an increase in the research of different fall de-
tection systems. Fall detection systems can be grouped
into the following categories: camera-based, ambient
sensors, and wearable sensors. The detection algorithm
and the sensors used can affect the accuracy of the sys-
tem. The detection algorithm used can either be a deci-
sion tree or machine learning algorithms. In this paper,
we study the different fall detection systems and the
problems associated with these systems. The fall detec-
tion model which most recent studies implements will
be analysed. From the study, it is found that personal-
ized models are the key, for creating an accurate model
and not limiting users to specific activities to perform.
Keywords Fall detection ·Machine learning ·Elderly
care ·Healthcare
1 Introduction
An increase of more than 30% in the elderly popula-
tion is expected by 2050 in 64 countries stated by the
United Nations [1]. The World Health Organization,
P. Vallabh
E-mail: phvallabh@gmail.com
R. Malekian
E-mail: reza.malekian@ieee.org
reported that about 28% of people aged 65 fall and
about 32% of people aged 70 fall each year [1], [2]. Due
to the shortage of nursing homes, more elderly people
are required to stay at home [3]. Elderly people who
live alone cannot alert anyone for help if a fall occurs
due to any serious injuries sustained or if they were un-
conscious [1]. The definition of a fall is as follows, “an
event which results in a person coming to rest unin-
tentionally on the ground or other lower level, not as
a result of a major intrinsic event (such a stroke) or
overwhelming hazard” [4]. A fall can occur in one sec-
ond usually it takes between 0.45 and 0.85s; during a
fall, the posture and shape of the person changes [5].
These changes are of great importance when detect-
ing a fall [5]. The risks of fall can be divided into two
categories, namely, extrinsic, and intrinsic risks [6], [7].
Extrinsic risks are related to environmental factors such
as drug usage, slippery floors, poor lighting, loose car-
pets, unstable furniture, clutter, and obstructed paths;
whereas intrinsic risks are related to the characteristics
of the person such as age, general clinical condition,
mental impairment, sedentary behaviour, impaired mo-
bility and gait due to reduced muscle strength [6], [8],
[7]. Extrinsic factors can be prevented by taking precau-
tions, whereas intrinsic factors cannot be prevented [7].
Factors that contribute to the increase in rate of falls
is the increase in person age, mortality, morbidity, dis-
ability, and frailty [9], [10]. When a fall occurs, it can
result in serious damage to a persons health, fear of
falling (FOF), loss of independence, no social contacts,
lack of movements, and decrease in productivity, which
increases the risk of another possible fall [6], [2], [9],
[4], [11]. This can result in loss of self-confidence which
can lead to social isolation and lower the quality of life
[2]. An FOF is linked to an increase of neuroticism and
anxiety which results in elderly people avoiding partici-
2 Pranesh Vallabh, Reza Malekian
pation in any physical activities [2]. The biggest danger
of falling is “long-lie” condition, where the fall victim
is unable to stand up from a fall and remain on the
ground for hours [12]. Long lie can result in dehydra-
tion, internal bleeding, physiological and psychological
consequence, depending on the seriousness of the in-
jury; and where half of the people who experience long
lie die within 6 months [6], [13], [14], [15], [16], [17].
With a fall detection system, the psychological stress
and severity of head-trauma during epileptic seizures
can be reduced and the cost of treatment is also re-
duced drastically [18], [5]. Falls can influence the in-
crease in the economic costs which impose a burden on
the health-care system [11]. A lot of research has been
done on fall detection systems from the 1990s [19].
The cost of being monitored is expensive at health
care facilities, where the main purpose of it is to de-
tect whether a person has fallen or not, that’s why fall
monitoring systems play an important role in society
by allowing people to be monitored from the comfort of
their homes or anywhere else resulting in huge savings
and eliminating the need for a 24/7 nursing to monitor
the person [14], [18]. The first fall detection system was
a device with a button known as user-activated devices
or Personal Alarm System (PAS), which was usually
worn as a wrist band or necklace and it required the
user to be conscious when a fall had occurred to press
the button and alert the emergency personnel [6], [20],
[3]. The problem with the push buttons were that they
could not be pressed if the user had lost consciousness
or was in a confused state due to panic; and the but-
ton could also be accidentally pushed; and the device
is not have been worn by the user during a fall [6], [21],
[17]. An automatic real-time activity recognition device
that can successfully discriminate between activities of
daily living (ADL) and fall activities is required. ADLs
contain a wide set of actions characterizing the habits
of people, especially in their living places e.g. walking,
sitting, standing, etc. [22]. These fall detection devices
that are available in the market are not satisfactory in
terms of high false alarms, high maintenance cost, and
they are not ergonomic [7]. Fall detection needs to de-
tect quickly to reduce impact and recovery time; and
should inform others quickly to reduce the time peo-
ple remain on the floor and to neglect any injuries that
can occur [20], [4], [23]. A precise, robust, and reliable
fall detection system is required for elderly people liv-
ing independently thus reducing the risks when living
alone [19], [20], [23]. There is no standard method for
fall detection in terms of what type of sensors that can
be used, which features to extract, and which machine
learning algorithm performs better [9]. The following is
expected from a fall detection system: no intrusion on
the users privacy, no restrictions on the users indepen-
dence, and should not degrade the users quality of life
[7].
There are several fall detection surveys published
which cover some aspects of the fall detection model. In
[2], an overview of wearable sensors is provided; partic-
ularly fall detection systems which incorporates smart-
phones are covered. The study also conducts an experi-
mental testbed to analyse the performance of the differ-
ent threshold fall detection algorithms that make use of
accelerometer sensors [2]. The results from the testbed
indicate that accelerometer techniques for identifying
falls are strongly influenced by the fall patterns; and
the tests also shows that it is difficult to set an acceler-
ation threshold to achieve high accuracy [2]. In [24], a
detailed comparison of wearable fall detection devices
and fall prevention systems are provided. This includes
the different sensors for detecting falls, the challenges
and design issues faced are discussed. A short analy-
sis on camera-based and ambient sensing is provided
[24]. The general learning models which are employed
in wearable systems are explained and the most pop-
ular supervised machine learning algorithms are anal-
ysed [24]. A three-level taxonomy which describes the
risk factors that are associated with falls, is proposed
[24].
In [25], an overview and a comparison of the dif-
ferent systems that make use of acceleration methods,
methods that combine acceleration methods with other
methods, and methods that do not use acceleration. In
[26], a detailed review on context-aware systems and
wearable accelerometer fall detection studies are pro-
vided; which includes comparisons between the differ-
ent studies. The challenges in design of the fall detec-
tion systems and the issues which affect the systems
performance; the trends in the present and future of
fall detection systems are identified [26]. Due to the
lack of fall data, the problem cannot be solved by us-
ing supervised machine learning algorithms [9]. In [9], a
taxonomy is proposed for sufficient, insufficient and no
training data on falls. A comprehensive overview on the
different techniques that can be applied for sufficient
fall data and the lack of fall data is described. A review
on camera-based and wearable studies for anomaly de-
tection is provided [9].
This study will provide an updated overview on the
different types of fall detection systems and the prob-
lems associated with each of the mentioned fall de-
tection systems. The study provides in-depth analysis
on the different categories compared to previously re-
viewed papers. The need for personalized systems will
be investigated and how it can solve the key problems
faced in fall detection system.
Fall detection monitoring systems: A comprehensive review 3
2 Model of a fall detection system
In figure 1, the most common fall detection model which
is used in many studies, when designing a fall detection
system is shown. The model comprises of the follow-
ing parts which will be discussed below: data collection,
feature extraction, feature selection, classifier, and eval-
uation.
Fig. 1 Fall detection classifier model.
2.1 Data collection
Fall detection starts by the collection of data from sen-
sors. These sensors can be either wearable sensors or
ambiance sensor, and camera-based sensors. These sen-
sors will be discussed in more detail, in section 3.
2.2 Feature extraction
Feature extraction is a method where significant at-
tributes are found from the raw data which consists of
meaningless information; and it plays a vital part in de-
termining the accuracy of the fall detection system [5],
[24], [27]. Fall detection systems require a distinctive
feature to represent the different activities and needs
to be able to classify falls from ADLs [28]. There are
different features, each having relevant characteristics
to specific ADLs or fall activity being performed [27].
Features can be group into two categories, namely, time
or frequency based features [27]. In wearable device, the
most popular features are acceleration magnitude of the
accelerometer and angular magnitude of the gyroscope
[24]. In camera-based systems the aspect ratio is the
most common one; whereas in Doppler and acoustic
device the Mel-frequency cepstral coefficient (MFCC)
features are the most popular ones. A lot of features
are calculated using statistical models such as median,
max, min, variance, etc. [27]. Special attention should
be applied when selecting features to produce a small
descriptive dataset [24]. The dataset descriptive power
is impacted by the number of features that the dataset
is comprised of [24]. Extracting features are performed
on data using a sliding window method [27].
2.3 Feature selection
The more features a database has, the more descrip-
tive it becomes, and it becomes difficult to find mean-
ingful relationships among the classes as the feature
space grows exponentially; and the performance of the
machine learning algorithm is also dependent on the
feature space [19], [5], [24], [27]. By finding features
which describes the data better and discarding the re-
dundant features, we can improve computational speed
and prediction accuracy [19], [24]. The method of se-
lecting features from an N dimensional feature space
is known as feature selection [17]. The feature selec-
tion algorithms are used to detect and discard features
that provides minimum contribution to performance of
the classifier [27]. Feature selection provides the follow-
ing advantages it reduces the cost of pattern recogni-
tion process, reduce the dataset, and provides better
accuracy [19], [17]. There are two categories of feature
selection methods, namely, filter methods and wrap-
per methods. Filter methods or ranking method make
use of search algorithms to score the different features
and rank the features from the best to the worst [24],
[27]. Filter methods make use of statistical tests such as
T-test, F-test, Chi-squared, etc. The wrapper method
takes combination of different features and compare the
combinations of features based on the classifier results,
where the classifier is part of the selection process [24],
[27]. The combination of features is chosen based on
that which provides an accurate model for classifica-
tion [24], [27]. The disadvantage of wrapper methods
is that it requires a huge amount of processing power
and it is very time consuming [27]. Instead of selecting
features, all the features that are extracted are com-
bined to create new features using principal component
analysis (PCA). The PCA is an unsupervised linear
transformation method, which useful variable reduction
procedure widely adopted in many fields and is a com-
mon technique for identifying patterns in data of high
dimension and expressing the data in such way as to
highlight their similarities and differences [22], [1]. A
PCA algorithm provides an orthogonal transformation
of a large feature space, into a new set of values made
of linearly uncorrelated variables called principal com-
ponents which results in a significantly smaller feature
space and decreases in dimensionality [22], [1].
4 Pranesh Vallabh, Reza Malekian
2.4 Classifiers
The fall detection classifiers can be divided into two
parts, threshold-based or rule-based and machine learn-
ing algorithms [2], [18], [4].
2.4.1 Threshold or rule-based
The most popular classification method used in fall de-
tection studies is the threshold analytical method [9],
[29]. The basic principal for the threshold analytical
method is that a possible fall could be detected based
on the sensors captured value; which is compared to
the reference value [9], [29]. A threshold method is a
flowchart where each node is tested where the outcomes
result in each branch. Fall detection that make use of
accelerometer sensors, uses a threshold parameter to
detect falls such as absolute acceleration magnitude or
wavelet acceleration sum-vector and compares it to a
predefined value [1]. The predefined value is calculated
and determined from a fall signal [1], [18]. Fall training
data is required to compute the threshold value using
domain knowledge’s or data analysis techniques [9]. The
advantage of threshold is that it is easy to implement,
power budget, and computational power [22], [2], [29].
The problem of threshold systems is that they lack lim-
ited recognition ability, not precise enough, difficult to
determine the predefined value and it results in high
false rates from running or jumping which results in
low accuracy [1], [29], [8]. The performance of the fall
detection methods is affected by the selection of the
fall indicators and detection thresholds [30]. Thresholds
results in low accuracy which makes researchers to fo-
cus more on machine learning classifiers which achieves
higher accuracies.
2.4.2 Machine learning
Classifiers obtained a greater performance compared to
threshold classifiers when using an accelerometer sensor
[18]. Machine learning algorithms have complex imple-
mentation when compared to the threshold implemen-
tation, it is based on decisions using posture calculation
which result in a higher fall detection rates [29]. The
advantage of machine learning algorithm is that the
different falls could be customized; and high accuracy
is achieved when compared to the threshold methods;
and it can manage anomalies (such as noise and incom-
pleteness) well; and it can detect patterns in signals [4],
[29]. The disadvantage of machine learning algorithm is
that it requires huge amounts of representative training
data, it is complex and requires heavy processing [4],
[29], [31]. Machine learning algorithm can be divided
into two groups supervised and unsupervised learning
algorithm.
2.4.2.1 Supervised
Supervised learning algorithm make use of labelled data
for training the system and the outputs of the system is
controlled [24], [27]. Certain classifiers can perform bet-
ter on certain activities [27]. Classifiers can be combined
such as voting machines or comparator machines [18].
A hybrid framework which make use of both threshold
based and machine learning algorithms is implemented
in the study [4]. Popular supervised machine learning
algorithms include Naive Bayes, k-Nearest neighbour,
support vector machine, hidden Markov model, and ar-
tificial neural network.
For a k-Nearest neighbour (k-NN) also known as a
lazy learner, which classifies a new feature vector based
on the classes of the other training feature vectors [7],
[10], [24]. Each time a new feature vector is inserted
into the classifier, all the training feature vector sets
are compared to the new feature vector in terms of
Euclidean distance. From the Euclidean distance, the
shortest distance will be determined, what centroid the
feature vector has joined and in what class it lies [18],
[10], [24]. The value k determines the number of cen-
troids that are available for each class. Special atten-
tion should be applied to determine the value of k; if a
smaller value k is selected the variances increases and
the results are less stable; and a large k value will re-
sult in an increase in biasing which will reduce the sen-
sitivity [7]. The disadvantage of this classifier is that
the time complexity increases as the training data in-
creases.
The Support Vector machine (SVM) uses a ker-
nel trick as it transforms the inputs, which are fea-
tures extracted, into a higher dimensional space using
a non-linear mapping in which an optimum hyperplane
is found separating two classes from a given training
dataset [15], [32], [33]. The basic idea is to find a sepa-
rating hyperplane that corresponds to the largest pos-
sible margin between the points of the different classes
[34]. A hyperplane is used to separate the two classes
by creating a decision boundary (maximum margin hy-
perplane) [24]. Optimization of separating hyper plane
is done by maximizing the distance between the hyper-
plane and the nearest data points [34], [32]. The maxi-
mum margin hyperplane is learnt based on the support
vectors, which the classifier uses to classify the new fea-
ture vector [15], [24].
A Hidden Markov Model (HMM) is a statistical
Markov model. An HMM, is made up with different
number of states. A typical model for a fall detection
Fall detection monitoring systems: A comprehensive review 5
system is a continuous HMM model, where each state
is connected to one or more states. An HMM consist of
the following parts: a transition probability distribution
matrix which is used to determine the probability of one
state reaching another state in one single step, an ob-
servation symbol probability distribution matrix which
is used to determine the output of a state based on the
input feature and an initial state distribution matrix
which is used to determine what the initial state is.
The system is trained by using a Baum-Welch training
algorithm. The class is determined using a Viterbi al-
gorithm [35]. The disadvantage of HMM it is computa-
tionally expensive and requires many model parameters
[4].
2.4.2.2 Unsupervised
Unsupervised learning algorithm make use of unlabelled
data for training the system [24]. This type of learning
algorithm can be trained on only fall data or non-fall
data [9]. The classifier can be trained on with new ac-
tivities on the fly. Popular unsupervised classifiers in-
clude: one class support vector machine, and nearest-
neighbour.
One class support vector machine (OCSVM) con-
verts the data to a feature space which is surrounded
by a hypersphere; and it searches for the appropriate
hyperplane that splits a portion of the input data from
the rest of the data by the sign of the distance to the
hyperplane (f(C) >0 or f(C) <0) [36], [37], [38]. The
classifier makes use of hyper-plane as a decision bound-
ary to classify the binary data [36]. The advantage of
OCSVM is that it describes the data in a flexible way;
since it does not need to ensure that the data follow a
certain distribution [14], [38].
Nearest-neighbour (NN) is a data driven method,
and which is simply a k-NN classifier where k is equal
to 1[15], [39]. The basic concept of NN is to allocate
the incoming record to the class that has a record clos-
est to the incoming record [39]. The Euclidean distance
is computed for the incoming record with each of the
stored record, where the minimum distance between the
incoming record and stored record is used [15], [37]. If
the minimum distance is higher than a threshold value,
the incoming record is considered an anomaly [15], [37].
The performance of NN will suffer if the data has re-
gions of varying densities [37].
2.5 Testing and evaluation of the system
Typical testing of the system is to perform leave-one-
out method or cross validation method [4]. The dataset
can also be split into 70% for training the classifier
and 30% for testing the classifier [27]. Statistical tests
are done to determine the overall performance of the
classifier [24]. The classification model can produce the
following four possible outcomes [18]: 1. True Positive
(TP) when a system properly detects a fall when fall
has occurred. 2. False Positive (FP) when a system de-
tects a fall when no fall has occurred. 3. True Negative
(TN) when a system detects no fall when no fall has oc-
curred. 4. False Negative (FN) when a system detects
no fall when a fall has occurred. False negatives are falls
which remained undetected and false positives are ADL
activities which were classified as falls [2]. The follow-
ing below, are most popular methods for measuring the
performance of the classifier.
The recall or sensitivity measures the ability of a
fall detection algorithm to correctly identify falls over
the entire set of fall instances [18], [11], [24].
recall =T P
T P +F N (1)
The precision measures the ability of a fall detection
to correctly identify falls over the entire set of instances
classified as falls. The precision measures the ability of
the classifier to return the fall results were correctly
classified [18], [24].
precision =T P
T P +F P (2)
The specificity measures the ability of a fall detec-
tion algorithm to correctly identify ADLs over the en-
tire set of instances classified as ADLs [18], [12], [11],
[24].
specificity =T N
F P +T N (3)
Accuracy is measured the portion of fall results that
were correctly classified amongst all outcomes [18], [24].
accuracy =T P +T N
T N +T P +F P +F N (4)
The F1-measure combines the precision and sensi-
tivity indicators [16], [24].
F1−measure =2×precision ×recall
precision +recall (5)
The receiver operating character (ROC) theory has
been used to properly define threshold values based
on constraints on the system sensitivity and specificity
[22]. By adjusting the threshold value, the ROC curve
is created [15]. From the curve, the threshold point is
selected where the maximum geometric mean of the
sensitivity and specificity is selected from equation 6
[15].
geometricmean =pspecif icity ×sensitivity (6)
6 Pranesh Vallabh, Reza Malekian
The area under the curve (AUC) is the recover op-
erating characteristic (ROC) curve and tells the perfor-
mance of the classification model [40], [41]. The closer
the AUC is to 1 the better the performance of the clas-
sification model is.
3 Fall detection sensors
Fall detection systems are also known as context-awareness
systems should be able to recognize, interpret, and mon-
itor different activities the user performs and be able
to detect fall events [7]. There are different types of
fall detection methods which includes camera-based,
acoustic-based, and wearable sensors [11]. Each method
of fall detection consists of numerous sensors, but none
of these sensors provides 100% accuracy, but each sen-
sor has its own advantage [42]. Table 1, shows the gen-
eral characteristics of these sensors types.
3.1 Wearable sensors
Due to the increase in wearable telemedicine technol-
ogy, solving these problems becomes easier [10]. The
growth of Micro-Electro-Mechanical System (MEMS)
resulted in miniaturized, more compact, and low cost
[7], [43]. They can be easily integrated to other available
alarm systems in the vicinity or to the accessories that
the person carrier e.g. smartphones or smart watches
which can achieve a kind of non-intrusive and non-
invasive diagnosis and monitoring [7], [10], [44], [45].
The wearable sensors are connected to the subject of
interest (SOI) [5].
Wearable devices make use of embedded sensors to
calculate the motion of the monitored body in any un-
supervised environment, period of inactivity, and the
posture of the person [21], [2], [5]. The first automatic
fall detection system is a wearable device that is placed
on the user to detect falls which make use of accelera-
tion or rotation information [46]. Wearable sensors can
detect a fall by analysing the impact of the body with
the ground, and taking the body orientation post and
prior to a fall has occurred [47]. Wearable sensors are
not affected by the environment or by privacy concerns
[11]. Collecting activity data from wearable sensors is
not restricted to laboratory environment, which allows
collection of real world activities [12]. Wearable device
can be implemented using micro-controller or smart-
phones.
3.1.1 Using smartphone for activity monitoring
Smartphones are now equipped with MEMS sensors
which can be used to perform unobtrusive fall detection
monitoring; and smartphones are already integrated in
the daily life of users [22], [2], [9], [43]. The increase
in growth of technology has made smartphones more
popular and more commonly used than any specific fall
detection equipment, they are non-invasive, portability,
cost-effective, easy to carry; and work both indoors and
outdoors [2], [9], [48]. Figure 2 shows a list of different
high precision sensors that are nowadays available on
the smartphone. The biggest advantage of smartphone
is that it has most of these sensors integrated into it,
which does not require no extra device [22], [1]. The
biggest problem of smartphone devices used in fall de-
tection is the fact that the devices lack battery draining;
and have limitations in memory and real-time process-
ing capabilities [1], [2].
Fig. 2 Available sensors on smartphone devices.
3.1.2 Different types of wearable categories
Wearable fall monitoring systems are grouped into three
groups: fall alert, fall risk assessment, and impact pre-
vention [4]. Fall alert or PERS (personal emergency re-
sponse system) is implemented to alert medical per-
sonnel or caregivers to provide assistance to user in an
event of fall [4], [11]. Fall risk assessment is the study of
fall in terms of the cause of it, and detecting which pa-
tients should be monitored based on their movements
[4]. Impact prevention or FIPS (fall injury prevention
Fall detection monitoring systems: A comprehensive review 7
Table 1 Characteristics of different fall detection methods
Method Price Continuous
monitoring
Battery
problem Obtrusive Privacy
Monitor
multiple
people
Easy
setup
Affected
by the
environment
Wearable Cheap Yes Yes Yes Yes No Yes No
Ambient Medium No No No Yes No Yes Yes
Camera Expensive No No No No Yes No Yes
system) is used to detect a fall event before it hap-
pens, and triggers a protection or prevention device to
protect the user [4], [11]. An Example of FIPS is the
detection of falls in the pre-impact phase where an ac-
tivate protection devices can be used, such as an in-
flatable airbag or other projection device, to avoid any
injuries from the fall [30], [49]. PERS prevents a long-
lie by notifying caregivers when a fall is detected, since
some falls are too hard to get up from or the user is
in an unconscious state [11]. PERS is the most popular
type of system and more research being conducted into
it. PERS can be split into posture and motion devices
[47]. Only PERS system will be analysed, and not FIPS
as it relies on pre-fall data to detect a possible fall and
is shown to achieve a low accuracy in [4].
3.1.3 Different wearable sensors
Wearable sensors include but are not limited to tilt
switches, accelerometers, gyroscopes, pressure sensors,
magnetometers, and microphones [4]. Each sensor has
different characteristics and can operate independently
or in conjunction with each other.
3.1.3.1 Accelerometer
Accelerometer sensors are the most popular and widely
used sensors for detecting fall accidents and sensing
body motions; as it has high accuracy, even in noisy
measurements a well-read acceleration measurement down
to 0Hz [5], [11], [36], [42], [43]. Accelerometers are fea-
sible, effective, fast, easy to set up and operate, simple,
lightweight, low-power, and cost-effective solutions for
fall detection systems [20], [18], [50]. In [25] a study was
conducted to detect what type of wearable sensor can
accurately detect falls based on sensors that use accel-
eration, acceleration integrated with other sensor meth-
ods, and no acceleration sensors. The study concludes
that using sensors which can sense accelerations are
good at detecting falls; whereas methods that did not
use acceleration are less accurate and can lead to many
false alarms [25]. Falls can be detected by applying
different signal evaluation techniques on accelerometer
data [1]. The most popular feature extracted from the
accelerometers is the Signal Magnitude Vector (SMV)
which is given below,
SM V =px2+y2+z2,(7)
where x,yand zare the acceleration values along the X,
Y, and Z axis of the accelerometer [45], [25]. A fall ac-
celeration signal comprises of peaks and valleys, and fall
activities usually associated with large SMV peaks [42],
[47]. Fall decision which make use of only SMV and con-
siders only the abrupt peaks in the acceleration which
result in high FP, due to the sudden movements which
occur when performing complex movements, such as sit-
ting down fast, and jumping [2], [47]. Most acceleration-
based studies use a threshold-based algorithm for de-
tecting a fall which result in high false alarms, in order
to reduce false alarms, machine learning algorithms can
be implemented [43].
The placement of sensors also plays a vital role as
it can directly impact the accuracy of the fall detec-
tion techniques [25]. In [51] different positions on the
human body is tested to identify the best position for
the accelerometer. The following positions were tested:
head, waist, and wrist to detect falls [51]. The acceler-
ation information measured was compared to a thresh-
old to detect a fall [51]. The results show that place-
ment of the accelerometer sensor on the person head
and waist achieves a sensitivity of 97-98% and speci-
ficity of 100% when using a simple threshold algorithm
[51]. Investigation in [4], to determine what phase of
a fall and placement of the tri-axial accelerometer on
the body will achieve the best accuracy. Hybrid frame-
work which make use of rule-based knowledge and a
two-layer Gaussian classifier was implemented [4]. The
following accuracies were obtained at different phases
of a fall: 86.54% for pre-impact, 87.315% for impact,
and 91.15% for post-impact [4]. The paper found that
the side of the waist is the best position for the sen-
sor during post-impact, followed by head, wrist, and
front of waist, thigh, chest, ankle, thigh, and upper arm
[4]. The reason for not achieving 100% accuracy in the
post-impact phase include signal loss, post-impact and
high impact ADLs were classified incorrectly [4]. If falls
are analysed during post-impact phase, the chest is not
suitable placement since the data transmission path of
8 Pranesh Vallabh, Reza Malekian
an alert signal could be blocked by the user’s body [4].
The following sensors placements result in false posi-
tives by not being able to differentiate falls among sit-
ting and standing: head, upper arm, wrist, ankle, and
chest [4]. Placing the sensor close to the person centre
of gravity makes the sensor less sensitive to spurious
movements.
The disadvantage of accelerometer sensors is prone
to elevators and high-speed cars or trains [20]. The out-
put of the accelerometer does not only consist of accel-
eration but also gravity, which can create errors when
calculating the angles resulting in high false positives
[10]. Accelerometer systems lack the adaptability to-
gether with insufficient capabilities of context under-
standing [34], [12]. Accelerometer methods require high
sampling rate, which can result in fast battery drain-
ing [34]. In [52] it was investigated that threshold based
algorithms implemented on smartphone suffers a limi-
tation from the accelerometers. The assumptions from
smartphone fall detection system is that the hardware
sensors measure acceleration with sufficient precision
which is not the case [52]. The sensors from differ-
ent manufactures record values in significantly differ-
ent ranges for identical test sensors, which makes it
impossible to set a reliable threshold value [52]. The
accuracy of the system increases when accelerometer
is incorporated with other sensors such as gyroscope,
magnetometers, and barometers, the accuracy of the
system increases [1].
3.1.3.2 Gyroscope
The most common feature extracted from the gyro-
scope sensor is the magnitude of the resultant angular
velocity(w), which is given below,
w=qw2
x+w2
y+w2
z,(8)
where wx,wyand wzare the angular velocity along
the X, Y, Z axis of the gyroscope [10]. There are limited
studies that only make use of gyroscope sensor to detect
a fall.
In [53], a study was conducted to understand the
use and the contribution a gyroscope sensor has when
classifying physical activities. Accelerometer and gyro-
scope data were collected and fed into different clas-
sifiers [53]. The study concluded that by adding the
gyroscope sensor to the system can improve the accu-
racy, the reason being that gyroscope data makes use of
the objects orientation which most activities consist of;
since the accelerometer only measures the linear motion
along specified directions [53]. There are a lot of stud-
ies which combines both accelerometer and gyroscope
together [29], [10], [54], [55].
The disadvantages of low cost gyroscopes are that
they suffer from time varying zero shifts. This intro-
duces significant errors when calculating the angular
acceleration and angular position, using differential and
integral operations [29], [10], [54]. If the noise is not re-
moved and the data is accumulating, the error can be
huge [29]. The Kalman filter algorithm with dynamic in-
formation of the target is required to remove the noise,
in order to estimate the angle [29]. The gyroscope is
also only available in higher grade smart phones [33].
3.1.3.3 Health sensors
In [56], fall is detected using electromyogram (EMG)
sensors; which measures the muscle control signals. When
a fall occur, there is a change in heart rate, which can
be used to detect a fall. In [45] an accelerometer and
cardio-tachometer is used to analyse and detect falls.
When a person falls down, the state of person heart-
rate can increase anxiety [45]. When a fall occurs the
heart rate can be used to detect how seriousness of the
fall is [57]. The disadvantage of using health sensors, it
difficult to place on, and they can interfere when per-
forming ADLs.
3.1.3.4 Wearable Camera
Compared to wearable sensors, wearable cameras pro-
vide a much richer set of data including contextual in-
formation about the environment, which includes anal-
ysis of a variety of activities including falls [20], [23].
The wearable camera system monitored is not limited
to confined areas, and it can extend to wherever the
subject may travel [23]. Wearable cameras do not af-
fect the privacy of the user since it only records the
surroundings of the user environments; and the system
processes everything locally on the device and noth-
ing gets transmitted anywhere [23]. In [20] the study
make use of a camera system is worn on the user waist,
which can provide continuous monitoring and is not
limited to certain areas as compared to static cameras.
Advantages of this system is that the privacy concerns
are removed as opposed to the static cameras [20]. The
wearable camera system uses edge orientations and his-
tograms to detect falls; which can work effectively both
indoors and outdoors, but it is highly invasive for sub-
jects [8]. The wearable camera records the surrounding
environment, which will make other people around the
user uncomfortable, as it will seem as it is recording
other people.
3.1.3.5 Ambient sensors as wearable sensors
Fall detection monitoring systems: A comprehensive review 9
Ambient sensors such as pressure sensor and micro-
phone can be attached on the user footwear to detect
falls [13], [58]. The advantage of attaching ambient sen-
sors on wearable items, it can provide outdoor moni-
toring, and it is not limited in coverage area; since it is
attached on the user. The disadvantage of the system it
is influence by the environment. In table 2, a summary
of the different wearable fall detection studies is shown.
3.1.4 Disadvantage of wearable sensors
3.1.4.1 Placement and intrusion
The major disadvantage of wearable devices includes in-
trusion, undesirable placement of device, neglect, or not
wanting to wear them, and inconvenience to the users
movement [7], [5], [32], [33], [40], [47], [66], [67]. Neglect
or forgetting to wear the device, can resulting a wear-
able device an ineffective solution [34], [21], [66]. The
undesirable placements of sensor on the user body, can
cause obtrusiveness, inconvenience and uncomfortable
when performing ADLs [34], [14], [21], [5]. Wearable
devices which are placed on the belt around the hip,
cannot be worn when changing clothes; and sleeping
which results in the inability to monitor when a person
is getting up from the bed [34], [51]. The addition of
extra sensors causes the user to feel uncomfortable and
lead to certain degree of inconvenience [33]. One solu-
tion, is to allow the user to choose the placement of the
device, and the device should perform on-body sensor
localization to detect the location of the device on the
user [55]. This will eliminate undesirable placements.
To make it convenient to the user, trouser pocket lo-
cation can be used for placing device [48]. Bathroom
has a high occurrences of falling down, which make it
difficult for a person to wear a device in the bathroom,
since these systems are affected by water, and make it
uncomfortable when bathing [17], [68].
3.1.4.2 Power
Wearable sensors are all battery powered, which means
it cannot be used when the device is recharging or bat-
teries will have to be replaced [7], [16], [67]. The battery
problem can be compensated by implemented using low
sampling frequency scheme together with a hierarchical
scheme methodology [1]. This will also reduce computa-
tional complexity of the system thus saving processing
time [1]. To make the system usable , a smaller number
of sensors is preferable on the user [4]. The advantage
of keeping the number of sensors to a minimum is that
it can cope with resource constraint issues such as bat-
tery power, storage, computational power, and network
bandwidth [4].
3.1.4.3 Hardware and software
Wearable device are limited to the hardware and soft-
ware [47]. Each smartphone device has fixed number
of sensors built in, to add more sensors, the smart-
phone is required to be upgraded. A basic sensor that
is available in all smartphones is the accelerometer sen-
sor. Compared to microcontrollers where the software
is fixed; the software of the smartphone can be up-
dated anytime. Smartphones can address the problems
of a low-power microcontrollers where classification al-
gorithms are constrained to limited memory and pro-
cessing power. Most microcontrollers systems imple-
ment only threshold classification, whereas smartphones
can implement machine learning algorithms.
3.1.4.4 Generates a lot of false positives
Wearable sensors generates a lot of false alarms when
performing daily activities, which can lead to frustra-
tion of users [34]. The reason for poor accuracy and high
false positives using accelerometers of lack of adaptabil-
ity with the lack of context understanding [44]. False
positives can be limited by implementing communica-
tion between the user and the device. If a fall has oc-
curred the user is communicated to first, to determine if
a fall has occurred. If the user does not respond within
a specified time period, the emergency service is com-
municated [69].
3.2 Ambient sensors
Ambient device make use of event sensing by collecting
and examine the environment which is used to track the
elderly person’s movement, through the use of externals
sensors which are attached around the surrounding en-
vironment such as a home or close to the subject [2], [5],
[8], [45]. Other application that ambient sensors provide
is indoor localization and security [5]. The advantage of
ambient devices is that user does not to need to wear
the device or remembering to put it on, it is passive
and unobtrusive [29], [47]. The ambient devices are non-
intrusive and it is invisible to the elderly which would
not affect the user privacy [19]. Ambient devices are
cheaper; but less intrusive compared to camera-based
systems [47].
3.2.1 Vibration detection
Ambient devices, that make use of vibration data where
the detection of falls is based on the characteristics of
vibration patterns [20], [47]. Vibrations can be used to
10 Pranesh Vallabh, Reza Malekian
Table 2 Summary of wearable sensors studies
Study Sensors Placement Features Algortihm Results
[47]
Smartphone compass,
accelerometer,
proximity,gyroscope
Trouser Pocket Mean, standard deviation, principal
component analysis
Decision tree,
support vector
machine
Accuracy: 90%
[18] Accelerometer Chest Wavelet coefficients Comparator
system Accuracy: 99%
[13]
Accelerometer,
pressure
sensor
Shoe Pressure, orientation of the foot State machine NA
[48]
Smartphone
accelerometer Trouser pocket SMV, Z axis High level fuzzy
Petri net Accuracy: 90%
[59] Electromyography
Lower limb
tibialis,
Gastrocnemius
muscles
Co-Contraction Indices Decision tree Sensitivity: 83.2%
Specificity:72.4%
[4] Accelerometer Waist
Mean, SMV, slope of SMV, standard
deviation Gaussian
mixture model Accuracy: 91.15%
[20] Smartphone camera Waist Gradient local binary patterns, edge
orientations Decision tree
Indoor accuracy:
93.78%,
outdoor accuracy:
89.8%
[11]
Accelerometer,
gyroscope, barometric
altimeter
Right anterior
iliac spine
SMV, orientation,height, downward
vertical velocity Decision tree Sensitivity: 80%
Specificity:100%
[10]
Accelerometer,
gyroscope Vest w, SMV k-NN Sensitivity: 95%
Specificity: 96.67%
[33] Smartphone compass
accelerometer, Trouser pocket Tilt angle, wavelet coefficients,
SMV
Support vector
machine, state
machine
Sensitivity: 92%
Specificity: 99.75%
[45] Accelerometer,
cardio tachometer
Wrist and
waist SMV, trunk angle, heart rate Decision tree
Accuracy: 97.5%
Sensitivity: 96.8%
Specificity:98.1%
[60] Accelerometer Waist
Signal magnitude area, tilt angle,
SMV Binary structure
classifier Accuracy: 95.6%
[61] Accelerometer Waist Sum of X and Z axis, total sum of X, Y,
and Z axis
One class
support vector
machine Accuracy: 96.7%
[62] Gyroscope Waist Pitch, roll, angular velocity Decision tree Specificity:100%
[63] Accelerometer Jacket collar
Inclination velocity, trunk inclination
angle Decision tree Sensitivity: 98%
[64] Accelerometer
Waist, neck,
right hand, left
hand
Inclination angles of X,Y, Z axis, SMV,
activity signal magnitude area Decision tree Accuracy: 92.92%
[65] Accelerometer,
barometric pressure Waist SMV, signal magnitude area, tilt angle,
differential pressure Decision tree
Accuracy: 96.9%
Sensitivity: 97.5%
Specificity:96.5%
Fall detection monitoring systems: A comprehensive review 11
detect fall based on an observation that normal activ-
ities cause measurable vibrations on the floor, which
means a when a user falls the down, the impact cause
by the body parts with ground will generate vibrations
that will be transmitted throughout the floor [70], [71].
An assumption is also made that the vibration signal for
falls and ADL are different [71]. Using the events and
changes in vibration data make it useful for monitoring,
tracking and localization [47]. Vibration signal can be
obtained using a piezoelectric sensor or an accelerom-
eter sensor. Floor vibrations are inexpensive, and they
can preserve the privacy of the user, but the perfor-
mance is influenced by the floor type and has a limited
detection range [39].
3.2.2 Acoustic detection
The basic idea of acoustic sensor is to make use of a mi-
crophone sensor to capture the movements of the users
where MFCC features are extracted to detect falls. The
MFCC features are extracted by first removing the high
frequency component [14]. Segmentation of the audio
signals into different frames [14]. A FFT transform is
applied to each frame to get the frequency spectral fea-
tures [14]. After the FFT, mel-scale mapping is per-
formed and finally discrete cosine transform is applied
to obtain 12 MFCC [14]. Applying beam-forming tech-
nique on the sound signal can enhance the desired signal
and reduce the interference from TV, radio, or phone
ringing [39]. Acoustic system makes use of a Rescue
Randy doll for mimicking human falls, for testing the
system [16]. The source of the sound signal, from mul-
tiple microphone can be detected using the steered re-
sponse power with phase transform technique, which
can work in any conditions [39]. The sound signal is
enhanced using beam-forming technique [39]. Classifier
design for acoustic fall design is difficult to design since
it is impossible to obtain realistic fall sound signatures
for training and testing of the system [72]. Generat-
ing fall data is difficult to simulate [72]. When captur-
ing simulating falls, the test subject tries to prevent a
painful fall [72]. Most of the acoustic studies make use
of Randy Rescue dolls which makes detecting low im-
pact falls difficult, to compensate these different weights
of rescue randy dolls is needed to train the system [68].
The studies which make use of Randy Rescue dolls can-
not replicate realistic falls sounds due to the hard skin
and the lack of bones in the mannequins [72]. The ma-
terial of the floor and the limited range of the detection
of the audio affect the system[16].
3.2.3 Pressure sensor
Pressure sensors are most common method for ambient
sensor since its low cost and non-obtrusiveness, a fall is
detected based on sensor pressure changes [5]. Pressure
senor used to detect the high pressure of the object due
to the objects weight for detection and tracking [5]. The
pressure changes depending on how close the person is
to the sensor [73]. If the person is closer to the sensor,
the pressure is high [73]. The disadvantage of pressure
sensors is the low detection precision which is below
90% [5]. The disadvantage of only using pressure sen-
sor to detect a fall, it can sense pressure of everything
in and around the object, which leads to false positives
hence low accuracy is achieved [5], [47]. The distance of
impact to where the pressure sensor is located can im-
pact the accuracy of the system [5]. Another problem is
that using only pressure sensors it cannot differentiate
between lying and falling postures [19]. To solve this
problem in [19], the make use of intelligent tiles which
consists of pressure sensors and three-axis accelerome-
ters. The accelerometer is used to detect hard human
falls, but cannot detect soft falls [19]. The accelerom-
eter is used to enforce the differentiation between the
falling and the lying down posture [19]. Each tile has
a processing unit and wireless connection and electric
power [19]. The disadvantage of the system is [19] is the
cost associated with each tile, and it requires power sup-
ply for each tile. Pressure sensors can have high false
alarms due to the fact the persons weight is not fac-
tored in, when detecting a fall; and the system is usu-
ally implemented on a small scale e.g. like a mat which
makes it costly when implementing it in a home envi-
ronment. The factors which influence pressure sensors
are the placement and sensitivity to pressure.
3.2.4 Passive infrared sensor
A Passive infrared (PIR) sensors detects falls using in-
frared signatures [35]. The strength of the received sig-
nal from the PIR sensors changes with motion of a hot
object within range of the sensors [74]. The PIR sen-
sor cannot be used to differentiate fall since a walking
person can produced a signal similar to a PIR fall sig-
nal [74]. In [74], a combination of both PIR and floor
vibration sensors is used to detect a fall. The PIR sen-
sors is used to reduce the false alarms in the system by
detecting whether the vibration signal was caused by a
human, and by detecting the presence of the user [74].
A fall alarm is ignored when there is no motion in a
room [74]. The biggest problem of using PIR sensors is
the line of sight and coverage area.
12 Pranesh Vallabh, Reza Malekian
3.2.5 Doppler sensor
Doppler sensors is a motion sensor that can sense, track,
and recognize moving objects and surveillance human
activity [41]. Doppler sensors are small and cheap which
only detects moving targets by suppressing stationary
background cluster, and are noise tolerant systems [75].
A Doppler sensor has different irradiation direction which
is less sensitive to the movements orthogonal to the irra-
diation direction compared to moving in the irradiation
direction, it becomes sensitive [75]. A Doppler sends a
continuous electrometric wave signal at the carrier fre-
quency and gets back the reflected wave which has the
frequency shifted by the moving object [75]. The ve-
locity of the moving object can be determined through
the frequency shift within the detection range [75]. The
disadvantage of Doppler sensor is sensitive to motion
and can penetrate apartment walls [75].
3.2.6 Electric near field
A near-field imaging (NFI) system uses floor sensors to
detect falls [76]. The floor sensors detect the locations
and patterns of the user by measuring the impedance
with a matrix of thin electrodes under the floor [76].
When the NFI is detected, the locations of the elec-
trodes from the matrix is detected [76]. More sensors
will be required when the area in the environment in-
creases, hence increase in cost. False positives are gen-
erated if there are pets or occlusions available. In table
3, a summary of the different ambient fall detection
studies is shown.
3.2.7 Disadvantage of ambient sensors
3.2.7.1 Coverage
The ambient sensors work only indoors or where the de-
vice is confined to, dead spaces, suffer from blind spots,
has limited recording area, it can only monitor one per-
son and it can be an expensive setup [20], [1], [29], [16].
The limited recording area does not affect electric near
field and pressure sensors, but will be expensive to cover
a monitoring area. Most of ambient systems, assumes
that only one person is present in the monitoring room.
3.2.7.2 Noise
Ambient sensors are affected by the environmental in-
terference, background noise and by ambient noise [3],
[7]. Ambiance device can produce many false alarms
due to other falls cause by everyday ob jects [2]. Acous-
tic and vibration sensors can only work on certain floor
type. Movement sensors are affected by obstructions or
occlusions which can deteriorate the signal.
3.3 Camera-based methods
The advancement in computer vision and image pro-
cessing techniques can also be applied in fall detection
problems, where a camera sensor is used to monitor
the user behaviour and detect fall activities without
interfering with the users routines [2], [8]. Camera sen-
sors can record the users position and shape [5]. Using
computer vision to detect a fall can be difficult since
the human body is composed of several parts which
can move freely, which makes the process of identifying
and locating people more difficult [21]. To overcome the
problem, the current studies uses human parts which
can be detected such as the head, waist, or feet [21].
The advantage of camera-based methods is that there
is no intrusion on the users since these sensors does not
need to be worn or remembered to be worn, due to the
fact that the camera system is contactless; and it can
be used to monitor one or more people simultaneously;
and it can be used to detect falls in public areas [29],
[8], [40], [43], [44], [47]. Multiple people can be tracked
in a frame through segmentation and marking module
[5], [8]. Camera-based methods can be used to serve
for two purposes, fall detection and security monitor-
ing. Advantage of camera-based methods compare to
the other methods, it is more robust and it can accu-
rately detect falls and different ADLs; and it can verify
a fall remotely if a fall has occurred [34], [9], [8], [47],
[78]. Camera-based systems is best suited where multi-
ple people need to be monitored e.g. hospital rooms or
old age homes etc [32]. Cameras are included in home
and care systems which have multiple advantages over
sensors based devices such as, multiple events can be
detected simultaneously with less intrusion. Figure 3
shows how a camera system detect a fall.
Fig. 3 Operations of the camera system when performing
fall detection.
3.3.1 Camera sensors
Falls can be detected using a single RGB camera, 3D-
based method using multiple cameras, and 3D-based
Fall detection monitoring systems: A comprehensive review 13
Table 3 Summary of ambient sensors studies
Study Sensors Features Classifier Results
[74] PIR and vibration
Single-tree complex wavelet transform form
vibration signal, amplitude from PIR sensor Support vector machine Accuracy: 100%
[17]
Vibration and
microphone Shock response spectrum, MFCC Naive Bayes Sensitivity: 97.5%
Specificity: 98.6%
[39]
Circular
microphone array MFCC Nearest neighbour Sensitivity: 100%
Specificity: 97%
[41]
Doppler and
motion sensors MFCC Support vector machine AUC: 0.98
[68] Accelerometer
and microphone
Shock response spectrum, MFCC,energy of
vibration signal, length and energy of sound
signal Gaussian mixture model Sensitivity: 95%
Specificity: 95%
[14] Microphone MFCC
One class support
vector machine Accuracy: 90.63%
[70] Special Piezo
transducer Vibration signal Pattern matching
Sensitivity: 95%
Specificity: 95%
[77] Far-field
microphone
Perceptual linear predictive coefficients and
Gaussian mean supervectors Support vector machine Recall: 94%
Precision: 70%
[71] Accelerometer
Peak to peak value, average rectified value,
weighted average rectified value, duration of
the signal, fast Fourier transform Decision tree Sensitivity: 87%
Specificity: 97.7%
[73]
Piezoresistive
pressure sensor Differential voltage Decision tree Sensitivity: 88.8%
Specificity: 94.9%
[35] PIR Differential voltage Hidden Markov model Accuracy: 80%
[75] Doppler Power spectral density, MFCC k-NN Accuracy: 93.3%
[76] Electric near-field Number of electrodes step on, the longest
dimension, magnitude
Two-state Markov
chain Sensitivity: 91%
method using depth cameras [34], [5], [78]. The most
popular vision-based method is the RGB camera which
is the cheapest and easy to setup [5], [32]. Multiple cam-
eras are required to cover a large area which can be
solved using omni-directional cameras or a wide-angle
camera can be used [21], [43]. The wide-angle cameras
have a wide field of view lenses which can be used to
monitor large areas [21]. The problem of this type of
camera, the images produced are highly-distorted [21].
The camera lens has high radial distortion which needs
to be corrected before the calibration process starts [21].
Omni-camera can capture can capture 360 degrees view
in a single shot which compensates for the blind spots
[79]. The lack of depth information from RGB cameras
can lead to a lot of false alarms [34], [43], [34]. The
2D camera methods can cause misjudgements when a
there are more than 2 people in the frame [66]. A sin-
gle camera cannot extract features that characterizes a
3-D objects movement which creates a robust fall detec-
tion system, but this can be created from multiple RGB
cameras [5], [67]. Multi-camera systems construct a 3-D
object from back projecting multiple silhouettes where
features such as velocity is extracted for detecting falls
[67]. For multi-camera systems installation, calibration,
and synchronising of the cameras in the same reference
frames is difficult, time-consuming and the cost of the
system increases [5], [67]. The 3D techniques which are
implemented from RGB cameras are not automatic and
requires manual initialization. Appearance deformation
can occur as the result of 2D grey or colour images that
are the projection of 3D targets [5]. The colour cameras,
in a controlled environment achieve high accuracy, but
would not work in an uncontrolled environment where
the lighting and tracking of user is fully controlled [34],
[43].
14 Pranesh Vallabh, Reza Malekian
Depth information alleviates the problems where
users or objects do not have consistent colour and tex-
ture, but they need to occupy an integrated region in
the 3D space [44]. Depth camera allows a person to
be extracted from an image at low computational cost
[44]. Depth cameras can be used to calculate the dis-
tance from the top of the person to the floor [5]. Depth
cameras can perverse the privacy of the user, and the
light conditions do not have any effect on it [34], [5],
[32]. Depth images can be extracted in dark rooms us-
ing an infrared light [34]. Depth cameras also can be
used to solve occlusion problems and track key joints of
the human body [5], [66]. The different depth cameras
include stereo vision, Time-of-Flight (TOF), and struc-
tured light camera [46]. Stereo vision camera constructs
a depth image from two views of a scene [46]. The prob-
lem of this camera the systems needs to be calibrated,
computationally expensive, and fails when the picture
does not contain enough textures [46]. The system can-
not work in low light conditions, which can be solved
by integrating an infrared light to it, but the loss of
colour information can cause segmentation and match-
ing difficulties [46]. The earliest depth camera was the
time-of-flight 3D camera, but the cost of setup is expen-
sive, and it is restricted to a low image resolution [5],
[46], [66]. Time-of-Flight image can be used to obtain
partial volume information which returns precise depth
image compared to stereo vision cameras for tackling
occlusion problems [46]. The most popular depth sen-
sor is structured light camera which includes the Kinect
sensor [46]. The Kinect sensor is a low-cost device which
comprises of infrared laser-based IR emitter, an infrared
camera and an RGB camera [34]. A Kinect, makes use
of infrared light sensors to illuminate the objects in
front of it and an infrared camera to observe them in
invisible light, the fall detection can be done at any
time [44], [43]. A Kinect sensor, can track the body
movements in 3D unlike 2D [34]. A Kinect senor, can
be used for human behaviour recognition, and detect
a fall in 24 day-night cycle [44]. The Kinect sensor is
not affected by the external light conditions due to the
depth interference is done by making use of an active
light source [34]. The Kinect sensor does not require
calibration since the automatic extraction of the fea-
tures [34]. The limitation of the Kinect sensor is that
the sunlight interferes with the pattern-projecting laser,
which is not suitable for outdoors [34].
3.3.2 Background subtraction and user tracking
Background subtraction is performed to extract the
moving object from the image known as foreground seg-
mentation. Simplest background subtraction technique
requires an original image with no moving objects. The
current frame is subtracted from original image to ob-
tain the moving object. The disadvantage of this tech-
nique it does not take into account the lighting changes,
shadow changes, and the changes in background due to
short-term movements [80]. This can be solved using
a Gaussian mixture model background model or using
approximate median filter [40], [81], [80]. Morphologi-
cal operations can be applied to reduce the noise in the
background. The extracted object is tracked continu-
ously, until the object is out of the camera view angle.
3.3.3 Camera-based detection methods for fal l detection
The camera-based detection can be split into shape
change, inactivity, posture, and 3D head motion [2], [5],
[45], [47]. In table 4, a summary of the different meth-
ods used to detect a fall is shown.
Simple method for detecting a fall using 2D method is
to locate the person in the video, and draw bounding
box around the person [46], [67], [82]. Most common 2D
feature extracted, includes aspect ratio [40], [46]. The
aspect ratio is computed as the ratio of the width of
the bounding box around the extracted object and the
extracted object height [40]. A small aspect ratio means
the users posture is upright, whereas a high aspect ratio
means the user posture is lying down [40]. Ellipse pro-
vides greater information than the bounding box; such
as calculating the fall angle [5], [46]. The fall angle of
the user is the angle between the long axis of the bound-
ing ellipse and horizontal direction [40]. A small angle
represents that person has fallen [40]. The problem of
using a bounding box alone, it does not provide enough
information regarding the human motion, and the per-
formance of this technique relies on the camera view
angles [83]. Analysing aspect ratio can be inaccurate
due to the position of the person, camera, and occlud-
ing objects. The method of analysing a fall by placing a
bounding box around a person can be efficient only by
placing the camera sideways and the accuracy of the
system depends on the occluding objects. In table 5,
a summary of the different camera-based fall detection
studies is shown.
3.3.4 Problem of camera-based sensors
Camera-based methods accuracy is dependent on how
efficient and accurate the shape modelling methods used
are[47]. The problem of camera-based systems is occlu-
sions, light conditions, coverage, privacy, cost, and high
processing.
Fall detection monitoring systems: A comprehensive review 15
Table 4 The different types of camera-based fall detection methods.
Type Description Advantages Disadvantages
Shape
change
When the shape of person
will change from an upright
position to a falling position,
over a certain time period [28].
It can be implemented
in real time, simple to model,
and easy to compute.
Using 3D shape requires more computation
and more cameras which can become
unreliable. Most shape-analyse algorithms
make use of a few features which is not enough
to detect sub-fall actions. The accuracy is also
affected by the proximity of the shape
attributes [28].
Inactivity
Is when a fall is detected
within an inactivity period
of how long has the person
being in a lying state.
It is fast as it has light
computing load, and it can be
run on a small computing
device.
False alarms are generated since they rely on the
context information such as period and threshold
and require the person to lie on the ground
for long which can cause serious injury. Some
users get up from the fall rapidly, hence the static
activity after a fall has occurred is not
captured [4].
Postures
From the user body, the
joints are tracked to determine
the posture.
High accuracy is achieved, as
it correlates the observed video
sequences to the stored ladled
video sequence [21].
It is computationally very expensive [21]. This
method requires a huge database, to successfully
recognise the different postures; and it is hugely
affected by occlusions.
Head
tracking
Fall is detected if there is
an occurrence of large
movement is associated
with the head. Fall is detected
based on the head speed.
This method can help to avoid
occlusion problems, since the
head is always visible in the
scene [84].
The head speed is greater when sitting down
fast which generate false alarms, and the head
speed is less during slow fall period which results
in fall not being detected [85].
3.3.4.1 Occlusions
Occlusions is where a room contains furniture or ob-
jects placed between the person and the camera which
can create false positives. When elderly people moves
to a smaller residence, they tend to take all these items
with them resulting in the room being fill with these
items, which means the user is partially occluded when
moving around the room [40]. Image processing difficul-
ties arises when changes occur in the monitoring area
e.g. furniture’s being shifted around the room; these
changes can also affect the accuracy of the system [14],
[40]. To accomplish the bounding box the RGB camera
is required to be placed sideways, which can fail due to
occlusions [82]. To solve this the camera is required to
be placed higher in the room not to suffer occlusions
and to have a greater field view [82]. In this case, de-
pending on the relative position of the person, the field
of view of the camera, a bounding box will not be suf-
ficient to discriminate a fall from a person sitting down
[82]. To avoid occlusions some researchers placed the
camera on the celling, where 2D velocity of the person
is used to classify the person. The problem of velocity
in a 2D method becomes high when the person is near
to the camera, which makes the threshold for differen-
tiating falls from sitting down fast difficult to define;
and 2D methods also suffer from occlusion problems,
this can be easily solved using 3D vision systems [46],
[28], [82]. Monitoring the whole body can fail when the
elderly people who struggle to walk are assisted with a
walking aid such as a rollator or walking frame which
causes the lower part of body to be occluded by the sys-
tem; and when objects are being carried [40], [85]. Head
tracking can also be used to solve occlusion problems,
where objects cover the user [84].
3.3.4.2 Light
Camera system should be able to monitor the user in
any light conditions [14], [40]. The different light sources
at the homes such as sun light, fluorescent light, light
bulbs, TV-screen, and the different light intensities that
occurs during the day, can result in overexposures in
some parts of the image, and the quality of the im-
ages is influenced [34], [3], [2], [40], [78]. Overexposure
can be slightly compensated through careful placement
of the camera in the room [40]. The problem of fore-
ground extraction using traditional cameras it relies on
background modelling in colour image space, when in
reality it is affected by lighting conditions and shadows
[34], [43], [44], [67]. The use of colour-based shadow
detection algorithms can be used to improve the out-
put of the background subtraction algorithm; but these
algorithms rely on an assumption that if an area is cov-
ered by a shadow, only the brightness of the image is
16 Pranesh Vallabh, Reza Malekian
Table 5 Summary of camera-based studies
Study Sensors Detection method Features Classifier Results
[78] Depth
sensor Posture Joints Decision tree
Accuracy: 93%
Sensitivity: 94%
Specificity: 91.3%
[83] RGB Shape change Velocity and motion
Support vector
machine Accuracy: 93.38%
[44] Kinect and
accelerometer Posture Human pose and motion
Fuzzy interference
and decision tree Accuracy: 98.6%
[66] Kinect Shape change
Vertical motion event
and 3D centroid Decision tree Precision: 94.31%
Recall: 85.57%
[34] Kinect and
accelerometer Shape change V-disparity Support vector
machine
Accuracy: 98.33%
Precision: 946.77%
Sensitivity: 100%
[21] Wide angle
camera Shape change
Angle, size of the upper
body Support vector
machine Accuracy: 97%
[46] Kinect Posture
Height of the user,
body velocity Decision tree Accuracy: 98.7%
[86] RGB Inactivity
Vertical volume
distribution ratio Decision tree Sensitivity: 99.7%
Specificity: 99.7%
[82] RGB Inactivity
Coebased on
motion history, aspect
ration, orientation Decision tree Sensitivity: 88%
Specificity: 87.5%
[38] RGB Posture
Ellipse shape structure,
position information of
silhouette
One class support
vector machine Accuracy: 100%
[87] 3D Head tracking
Vertical and horizontal
velocity Decision tree Accuracy: 78.9%
[85] RGB Head tracking
Direction of the body
and the ratio of the
variances in x and y
direction
Gaussian
multi-frame Accuracy: 85%
[84] RGB Head tracking,
shape change
Ellipse shape, position of
the head, vertical and
horizontal projection
histogram
Multi-class
support vector
machine
Sensitivity: 90.27%
Specificity: 95.16%
[88] RGB Posture
Skeltons and centroid
context String matching Accuracy: 96%
[80] RGB Shape change
Movement coe,
orientation, aspect ratio Decision tree Accuracy: 90%
Fall detection monitoring systems: A comprehensive review 17
affected and there is no change in colour information
[40]. There is a high risk of falls occurring in low light-
ing conditions compared to normal illuminated condi-
tions [44]. To solve the problem of lighting conditions
for single cameras an active source of infrared (IR) light
can be installed along with the camera; but there will
no colour available due to the IR illumination for back-
ground modelling [67]. Colour information is not avail-
able in near-infrared night images, and colour images
that are available during daytime are not reliable [40].
Depth cameras can solve the lighting conditions, and
can work during both day and night [46].
3.3.4.3 Cost and high Processing
The cost of the infrastructure and installation of sen-
sor equipment’s is expensive. Image quality in reality is
much lower than the lab experiment setup, this can be
accomplished by installing a high-quality camera which
can result in high cost [40]. Camera-based systems re-
quire considerable computational power running real-
time algorithms [44]. One way of minimising the compu-
tational power is to integrate the camera based system
with an accelerometer [44]. Camera-based system only
starts processing when a possible fall is detected from
an accelerometer sensor [44]. An accelerometer sensor
is used to identify if a possible fall has occurred and
the camera system is used to authenticate a fall [43].
The frames are not processed instead there are stored
in a circular buffer, and only processed when a fall has
occurred [43].
3.3.4.4 Coverage
Camera based systems can only work indoors or where
the devices are confined to, which can create blind spots,
occlusions cannot be detected, limited field view, and
dead spaces are created [34], [3], [1], [2], [9], [47], [78].
Multiples cameras are required to be installed to solve
these problems and provide continuous monitoring, which
increases the cost of the system [40]. Wide angle camera
can be used to provide coverage of the room, but the
spatial resolution of the camera system decreases due
to the lens of the wide-angle cameras [40].
3.3.4.5 Privacy
The ethical issues that are associated with camera-based
methods includes confidentiality and privacy of the mon-
itored person, which makes it difficult to monitor a per-
son in the bedroom and bathroom [34], [2], [47], [78].
The problem of colour camera based systems is that
they contain facial characteristics of users which results
in privacy concerns, which can be addressed by captur-
ing low quality images, using depth images or image
processing technique such as silhouettes [44], [70], [67].
Even though privacy techniques are applied, people still
has a feeling of “being-watched” based on their percep-
tion of a camera system [43], [70]. Instead of capturing
the user, the environment scene can be captured like in
[20] and [23].
4 Personalization
Personal information can make the system smarter by
adapting the different parameters for different person
[79]. If different body postures are not learnt, high false
rate could be resulted [29]. Methods that make use of
thresholds are most popular and easy to implement,
and computationally inexpensive, but does not work
on different people, and does not provide a good trade-
off between false positives and false negatives [9], [30].
People have different types of body figures; whereas us-
ing the same threshold in fall detection algorithm will
not work for everyone or would not be optimal [79].
With thresholds is difficult to adapt the threshold to
new types of falls and makes it work on different peo-
ple [2], [9]. Falls of elderly people might last longer than
that of young people [28]. The values from the thresh-
old method is determined without using any theoretical
and/or experimental basis; and where the fall detection
model fail is that it cannot address inter-individual dif-
ference [30]. The basic idea behind personalization, is to
train the system using the user data, which will result
in higher accuracy.
5 Personalization
Personal information can make the system smarter by
adapting the different parameters for different person
[79]. If different body postures are not learnt, high false
rate could be resulted [29]. Methods that make use of
thresholds are most popular and easy to implement,
and computationally inexpensive, but does not work
on different people, and does not provide a good trade-
off between false positives and false negatives [9], [30].
People have different types of body figures; whereas us-
ing the same threshold in fall detection algorithm will
not work for everyone or would not be optimal [79].
With thresholds, it is difficult to adapt the threshold to
new types of falls and makes it work on different peo-
ple [2], [9]. Falls of elderly people might last longer than
that of young people [28]. The values from the thresh-
old method is determined without using any theoretical
and/or experimental basis; and where the fall detection
18 Pranesh Vallabh, Reza Malekian
models fail- is that it cannot address inter-individual
difference [30]. The basic idea behind personalization,
is to train the system using the user data, which will
result in higher accuracy.
5.1 Design of a personalized model
Classification can be trained using the user data or non-
user data or the combination of both user and non-
user data. The use of supervised machine learning al-
gorithm cannot be used to solve the problem, as the fall
data that is used are from simulated falls [9]. Since falls
are rare, supervised machine learning algorithms cannot
be used [9]. Supervised algorithms can classify known
classes which they are trained [9]. Supervised machine
learning algorithm requires the data to be label which
result in waste of time and effort [9]. Supervised clas-
sifiers cannot provide a person-specific solution for in-
dividuals [38]. Due to the lack few fall data, supervised
classification algorithms may not work as desired, the
following classification are needed over/under-sampling,
semi-supervised learning, cost-sensitive learning, and
outlier/anomaly detection [9].
A large dataset needs to be created for training the
supervised classifier which should contain data for dif-
ferent activities; if a person does not fit the dataset
e.g. if the person is obese a good performance could
not be obtained for the specific individual [38]. Su-
pervised learning algorithms require a balance dataset
with has equal misclassification costs for the different
classes [9]. When unbalance data is used to train the
algorithms, the algorithms fail to distinguish the char-
acteristics of the data, which result in low accuracies;
and their prediction tend to favour the majority class
[9]. The imbalance class can be handle by performing
cost sensitive-classification, where the cost of the clas-
sification problem is treated differently [9]. This can
be accomplished by adding a cost matrix to a cost-
insensitive classifier or by integrating a cost function in
the classification algorithm to generate a cost-sensitive
classifier [9]. A cost matrix of a fall detection problem
is defined, by getting the optimal decision threshold of
the classifier [50]. Cost-sensitive analysis can be per-
formed for fall detection using Bayesian minimum risk
or the Neyman-Person method [50]. This is calculated
by varying the ratio of the cost of a missed fall to a
false fall alarm to determine an optimal region of op-
eration using the ROC curve [50]. Generally, the ratios
are fixed and should not be dependent on the dataset
used [50]. The costs are unknown and are difficult to
compute [9]. In [40], the study make use of a weighted
SVM to compensate the imbalance of data of the falls
and normal activates from the camera. The weights are
determined using cross-validation and grid search max-
imizing the area under curve of ROC [40].
The lack of fall data could also be compensated us-
ing sampling techniques to generate fall data [9]. Fall
can be oversampled or the normal activity class can be
under-sampled to train a supervised classifier [9]. The
disadvantage of oversampling it can lead to over-fitting
if a lot a lot of artificial data points are generated and
do not represent a fall [9]. The disadvantage of under-
sampling it can lead to under-fitting it the normal ac-
tivities class is reduced to match the number of total
activities of falls [9].
Another approach is to apply temporal patterns which
can be used to describe and provide more information
on the events that the user performs[89]. The temporal
paths are used to recognize or predict future events that
the user may performed [89]. In [89], the system com-
bines the temporal extension of Fuzzy Formal Concept
Analysis (data driven) and Fuzzy Cognitive Maps (goal
driven) approaches for better decision making [89]. The
system recognizes the following events: tiredness, sleep-
ing, having breakfast, and having dinner [89].
Classifiers only require normal activities for train-
ing, which eliminates data imbalance between fall and
normal activities are known as unsupervised machine
learning algorithm [9]. The problem is that if the nor-
mal behaviour in not properly learned, the system can
result in large number of false positives, as a slight vari-
ation from a normal activities can be detected as a fall
[9]. The classifier needs to adapt and learn new activ-
ities in order to reduce the false alarm rate when de-
tecting falls [72]. The advantage of the unsupervised
approach, is that the classifier can easily adapt to new
data without worrying about data imbalances [72]. In
table 6, below show the summaries of systems which
make use of personalized models.
The basic personalization is customizing the thresh-
old based on personal characteristics such as height,
weight, etc. [79]. In [79] an Omni-camera is used to
record the activities, where a bounding box is placed
on the user [79]. The system requires a background
image, no user present in the background [79]. To de-
tect a fall the foreground is extracted by performing
background subtraction [79]. A fall is detected if the
bounding box aspect ratio is greater than pre-defined
threshold value [79]. The predefined threshold value is
customize based on the following personal information
height, weight, and electronic health history [79]. The
reason for the personal information is used to adjust
the detection sensitivity which reduces false alarms, and
provide more attention to the elderly person with spe-
cific needs [79]. The use of electronic health history is
to increase the detection sensitivity automatically if the
Fall detection monitoring systems: A comprehensive review 19
person experiences cardiovascular disease or if a fall ac-
cident has happened before [79]. In [90] a smartphone
system which is based on the user information’s such
as the ratio of height and weight, sex, age is used to
adjust the threshold value and sampling of the accel-
eration data. From the tri-axis acceleration sensor, the
direction of the three-axis was extracted [90]. The sys-
tem calculates SMV [90]. Based on the BMI, the user
age, and sex, the maximum and minimum threshold
from the acceleration and the sampling frequency de-
termined through the range of the personal information
[90]. Fall is classified based on the thresholds and the
system achieves a sensitivity of 92.75% and specificity
of 86.75% [90].
In [37], a study was conducted to compare person-
alised systems to non-personalised systems using a smart-
phone accelerometer. Three unsupervised methods were
implemented NN, OCSVM, and (Local Outlier Factor)
LOF; and one supervised method SVM [37]. The study
was divided into two stages, the first stage is deter-
mining which unsupervised method was the best; and
the second stage to determine how does personalized
perform on both the best unsupervised method and su-
pervised method [37]. The raw data of the three axes
of the accelerometers are fed into the classifiers [37].
From the first stage, it was found that NN outper-
form the rest of the unsupervised methods [37]. For
the second stage, the personalized model of the NN is
trained with the normal activities of the user; whereas
the non-personalized model is trained with the normal
activities of other people data [37]. The personalized
model of the SVM is trained with the normal activities
of the user and fall activities of other people; whereas
the non-personalized model is trained with both nor-
mal and fall activities of other people [37]. It was found
that both the personalized model, NN and SVM outper-
form the non-personalized model [37]. The personalized
SVM model achieved slightly higher geometric mean of
0.9764 compare to the personalized NN model of 0.9688
[37]. The NN model is better compare to SVM model,
the reason being it can adapt to new data, and it can
recognize more fall types.
Another approach is to adapt the classifier to ac-
cept new ADL data and re-train the classifier in or-
der to learn the user movements. In [42] a smartphone
tri-accelerometer sensor was used with a NN classifier;
where the capture magnitude acceleration data is com-
pared to the store ADL data from the smartphone. A
fall is detected when the difference between the stored
pattern and incoming pattern is high [42]. The new
ADL is added every time the system classifies the in-
coming data as ADL; where the old ADL record is re-
placed with new ADL [42]. To reduce processing power
and computational time, the system only classifies when
magnitude of the acceleration value is greater than 1.5g,
and if long lie occurs [42]. The advantage of NN clas-
sifier is that it easy to add new data, and it does not
require simulated falls for the training the system [42].
The simulated fall data was used only for testing the
classifier [42]. The disadvantage of the system is that it
cannot detect soft falls and it uses long lie. If a person
attempts to get up from a fall but fails each time during
the long lie period, the system would not detect a fall
event [42].
6 Discussion
High classification accuracy is reported in almost all
of the fall detection studies, but it was conducted on
limited number of subjects, fall types and activities [4],
[12]. The reason for simulated falls, it is extremely hard
to collect real-world elderly person fall data; since 30%
of elderly population over age of 65 years old fall at
least once per year [12]. Current fall detection studies
are only tested in controlled experiments where they
achieve high accuracy, but when placed in the real world
the accuracy of these systems decreases [20]. Studies
test the specificity of ADL through laboratory experi-
ments by the same subjects who generate fall data [12]
These data could be biased, since subjects are forced
to perform activities, which are typically spontaneous
[12]. The choice of the mattress to reduce the impact
of the falls to protect the volunteers from injuries, can
reduce the accuracy of the system when applied to the
real world [12].
It is difficult to compare the different fall detection
studies in a fair play since each study made use of they
own dataset from different conditions [15]. The problem
comes in when comparing a system since each study
validated they research on different data collection pro-
tocols, subject groups, and environment settings, hence
they cannot be directly compared to previous studies
[4]. The factor which influence the performance is the
number of training samples are used for training the
system [15]. The main problem of acceleration based
studies is that it is difficult to compare the different
studies; since that each research study make uses its
own dataset composed of simulated falls and ADL [15].
It is difficult to judge whether the results obtained from
these studies are influence by the dataset complied, and
it is impossible to make a comparison since the dataset
used in each study are different [15]. Since these de-
vices are required to be worn for long-periods or the
whole day a complete dataset is required compared to
fall detection studies where the dataset is limited [20].
20 Pranesh Vallabh, Reza Malekian
Table 6 Summary of personalized fall detection systems
Study Sensor
Method
of
personalization
Features Algoirthm
Results
before
personalization
Results
after
personalization
[79] Omni-camera
Changing the
pre-defined
threshold value
based on user
height, weight
and electronic
health history
Aspect ratio Threshold
tree Accuracy: 78% Accuracy:
90%
[37] Smartphone
accelerometer
Training the
system with only
personalized data
X, Y, Z axis
from the
accelerometer
Nearest-
neighbour
Sensitivity:94.15%
Specificity:93.84%
Sensitivity:96.65%
Specificity:97.15%
[37] Smartphone
accelerometer
Training the
system with only
personalized data
X, Y, Z axis
from the
accelerometer
Support
vector
machine
Sensitivity:96.48%
Specificity:95.73%
Sensitivity:97.97%
Specificity:97.34%
[69] Smartphone
accelerometer
Changing
the pre-defined
threshold
value based on
user height,
weight, and
level of the
activity
Signal
magnitude
vector,
time,
period,
angle
Threshold
tree NA NA
[31] Accelerometer
Adding new
records to the
system
randomly, and
adapting the
system based
on user
weight
and mobility
Filtered
acceleration
value,
energy of
acceleration
Threshold
tree NA Sensitivity:100%
Specificity:95.68%
[42] Smartphone
accelerometer
Adding new
records to the
system
Signal
magnitude
vector
Nearest-
neighbour
Area under curve:
0.969
Area under curve:
0.978
[91]
Magnetometer
,accelerometer
, gyroscope
Self-adaptive,
update the
threshold for the
user
Signal
magnitude
vector
Threshold
tree
Accuracy: 90.7%
Sensitivity:97.7%
Specificity:79.8%
Accuracy: 92.1%
Sensitivity:98.7%
Specificity:81.7%
[90] Smartphone
accelerometer
Changing the
pre-defined
threshold value
and sampling
frequency based
on user BMI, age
and sex.
Signal
magnitude
vector
Threshold
tree NA Sensitivity:92.75%
Specificity:86.75%
Fall detection monitoring systems: A comprehensive review 21
In [12] evaluation was conducted on real falls based
on accelerometer fall detection algorithms where 29 real
world falls were tested on. The result from the evalu-
ation show a reduce sensitivity and specificity values
compare to when conducted in an experiment environ-
ment to evaluate the effectiveness of the algorithms to
detect falls in real-life events [12]. The study achieved
average specificity of the algorithms is 83.0% and av-
erage sensitivity of the algorithms is 57.0% which are
much lower compared to the simulated environment
[12]. There is a huge number of false alarms generated
from the algorithms in a one-day monitoring period
which ranged from 3 to 85 [12]. The results obtained
from the study is to encourage researchers to take real-
ity activities into consideration [12]. The problem with
these studies is that they cannot work in the real world
since no training data for falls were used, and low ac-
curacy will be achieved since classifier cannot predict
a fall that it has never observed before [9]. Collecting
fall data is futile as it requires a person to perform a
real fall which can result in serious injuries [9]. About
94% of fall detection studies used simulated falls from
laboratory experiments for training the classifiers [92].
This shows that the difficulty in obtaining real fall data
[92]. Instead of real falls, artificial falls are collected in a
controlled laboratory environment, which does not rep-
resent an actual fall [9]. The advantage of artificial fall
it provides information of how falls is occurring, but
does not make it easier for detecting falls [9]. Classi-
fiers which use artificial falls as training data can re-
sult in over-fitting, which can cause poor decisions on
the actual fall [9]. The fall data are limited quantity
and suffer from ethic clearance [9]. To get accurate fall
data, a long-term experiments needs be conducted in
nursing homes using wearable sensors, ambient sensors,
or camera based methods [9].
The main problem of vision based the absence of
flexibility, as these systems are case specific where they
are designed and optimized for a certain situations or
scenarios [2]. Camera-based studies algorithms are eval-
uated from data collected from controlled environment,
optimal conditions such as perfect illumination, simple
scenarios or scenes, and falls are simulated by actors
[40]. The challenges found from real life data compare
to the simulated data is that the image quality is low
and falls are rare and vary a lot in terms of speed and
the nature of fall [40]. Most studies make use of sim-
ulated data, where the falls been recorded in artificial
environments and the person performing are young peo-
ple [40].
Each individual has different characteristics and mo-
tion patterns compare to people used in the training
data [30]. Another problem is difficult to detect all the
ADL since the classifier is required to be trained with
each type of ADL [72]. The classifier needs to adapt and
learn new activities in order to reduce the false alarm
rate when detecting falls [72]. It is difficult to detect the
different types of falls for the different people; since a
fall has different acceleration characteristics and mag-
nitude of acceleration has high variation among various
body types [23]. The phone placement differs from per-
son to person [23]. The limitations of current fall detec-
tion studies are the difference in the shape or strength
of measured signals if healthy adults or elderly people
wear the fall detector and if the falls are simulated or
real, with possibly relevant effects on the design and the
performance of the fall detection algorithm [11]. ADLs
such as lying down and sitting down can generate high
impacts which can be misclassified as a fall, for over-
weight users [4]. Falls with recovery and backward col-
lapses, where users end up in a sitting position result in
a misclassification [4]. An actual free fall is not created
due to the cautiousness of subjects, which results in not
a proper fall detection [20]. Even when safety precau-
tions are there, subjects are still too afraid to fall [20].
The occurrence of fall rate is low, which results in insuf-
ficient or no data [9]. Different types of fall can occur,
which makes it very difficult to model [9].
The solution, is to create a personalized system;
which adapts and learns the users movements. By learn-
ing the users movements, the system will be available
to recognize a wide range of ADLs and not force the
user to perform certain activities. One way to achieve
a personalized system, is by using unsupervised ma-
chine learning algorithm; which can easily adapt new
data without worrying about data imbalances [72]. Un-
supervised algorithm, would only be required to trained
with ADLs which are easier to capture compare to fall
activities. The biggest advantage of personalized sys-
tem, it will work on anybody, regardless of their weight
and height.
7 Conclusion
In this paper, the different fall detection systems that
exist were discussed and analysed, where each one has
their own advantages and disadvantages. The accuracy
of the system depends on the sensors used and the type
of classifications. The wearable and camera-based sen-
sors are the most popular ones compared to ambience
sensors. Ambiance sensors are highly influence by the
environment. The wearable sensor can include a device
of MEMS sensors or the use of a smartphones and the
system can include a false alarm button. Camera-based
sensors, main disadvantage is that the limited cover-
age and the performance being affect by objects in the
22 Pranesh Vallabh, Reza Malekian
environment. The wearable devices main disadvantage
is that it intrusive and the placing of the device on
the human body is uncomfortable. Wearable sensors
are preferred method as it is practical and allows for
continuous monitoring and is not influence by the en-
vironment. The wearable sensor also provides outdoor
monitoring, and can be used to collect real data in a
cost-effective approach. A smartphone can be used as a
wearable device since a lot of people have them, and it
is not intrusive. Wearable device can be placed in the
user pocket, which would not interfere when the user is
performing ADLs. Experimental systems are limited to
the laboratory setting, which would not work in reality
and is limited to certain ADLs. Personalization is key,
in fall detection system; since it does not only increase
the accuracy of the system, but can also be adapted
to learn new activities. Adapting new activities can be
done by implementing an unsupervised machine learn-
ing algorithm, since data balance would not be an issue.
References
1. de la Concepci´on, M. ´
A. ´
A., Morillo, L.M.S., Garc´ıa,
J.A. ´
A., Gonz´alez-Abril, L.: Mobile activity recognition
and fall detection system for elderly people using ameva
algorithm. Pervasive and Mobile Computing 34, 3–13
(2017)
2. Luque, R., Casilari, E., Mor´on, M.-J., Redondo, G.: Com-
parison and characterization of android-based fall detec-
tion systems. Sensors 14(10), 18543–18574 (2014)
3. Garripoli, C., Mercuri, M., Karsmakers, P., Soh, P.J.,
Crupi, G., Vandenbosch, G.A., Pace, C., Leroux, P.,
Schreurs, D.: Embedded dsp-based telehealth radar sys-
tem for remote in-door fall detection. IEEE journal of
biomedical and health informatics 19(1), 92–101 (2015)
4. Pannurat, N., Thiemjarus, S., Nantajeewarawat, E.: A
hybrid temporal reasoning framework for fall monitoring.
IEEE Sens. 17, 1749–1759 (2017)
5. Yang, L., Ren, Y., Zhang, W.: 3d depth image analysis
for indoor fall detection of elderly people. Digital Com-
munications and Networks 2(1), 24–34 (2016)
6. De Backere, F., Ongenae, F., Van den Abeele, F., Nelis,
J., Bonte, P., Clement, E., Philpott, M., Hoebeke, J.,
Verstichel, S., Ackaert, A., et al.: Towards a social and
context-aware multi-sensor fall detection and risk assess-
ment platform. Computers in biology and medicine 64,
307–320 (2015)
7. ¨
Ozdemir, A.T., Barshan, B.: Detecting falls with wear-
able sensors using machine learning techniques. Sensors
14(6), 10691–10708 (2014)
8. Yang, L., Ren, Y., Hu, H., Tian, B.: New fast fall detec-
tion method based on spatio-temporal context tracking of
head by using depth images. Sensors 15(9), 23004–23019
(2015)
9. Khan, S.S., Hoey, J.: Review of fall detection techniques:
A data availability perspective. Medical Engineering &
Physics 39, 12–22 (2017)
10. Jian, H., Chen, H.: A portable fall detection and alerting
system based on k-nn algorithm and remote medicine.
China Communications 12(4), 23–31 (2015)
11. Sabatini, A.M., Ligorio, G., Mannini, A., Genovese, V.,
Pinna, L.: Prior-to-and post-impact fall detection using
inertial and barometric altimeter measurements. IEEE
transactions on neural systems and rehabilitation engi-
neering 24(7), 774–783 (2016)
12. Bagal`a, F., Becker, C., Cappello, A., Chiari, L., Aminian,
K., Hausdorff, J.M., Zijlstra, W., Klenk, J.: Evaluation
of accelerometer-based fall detection algorithms on real-
world falls. PloS one 7(5), 37062 (2012)
13. van de Ven, P., O’Brien, H., Nelson, J., Clifford, A.: Un-
obtrusive monitoring and identification of fall accidents.
Medical engineering & physics 37(5), 499–504 (2015)
14. Khan, M.S., Yu, M., Feng, P., Wang, L., Chambers,
J.: An unsupervised acoustic fall detection system us-
ing source separation for sound interference suppression.
Signal processing 110, 199–210 (2015)
15. Igual, R., Medrano, C., Plaza, I.: A comparison of pub-
lic datasets for acceleration-based fall detection. Medical
engineering & physics 37(9), 870–878 (2015)
16. Principi, E., Droghini, D., Squartini, S., Olivetti, P., Pi-
azza, F.: Acoustic cues from the floor: A new approach
for fall classification. Expert Systems with Applications
60, 51–61 (2016)
17. Zigel, Y., Litvak, D., Gannot, I.: A method for automatic
fall detection of elderly people using floor vibrations and
soundproof of concept on human mimicking doll falls.
IEEE Transactions on Biomedical Engineering 56(12),
2858–2867 (2009)
18. Gibson, R.M., Amira, A., Ramzan, N., Casaseca-de-la-
Higuera, P., Pervez, Z.: Multiple comparator classifier
framework for accelerometer-based fall detection and di-
agnostic. Applied Soft Computing 39, 94–103 (2016)
19. Daher, M., Diab, A., El Najjar, M.E.B., Khalil, M.,
Charpillet, F.: Elder tracking and fall detection system
using smart tiles. Sensors 15800, 1 (2016)
20. Ozcan, K., Velipasalar, S., Varshney, P.K.: Autonomous
fall detection with wearable cameras by using relative
entropy distance measure. IEEE Transactions on Human-
Machine Systems 47(1), 31–39 (2017)
21. Bosch-Jorge, M., S´anchez-Salmer´on, A.-J., Valera, ´
A.,
Ricolfe-Viala, C.: Fall detection based on the gravity vec-
tor using a wide-angle camera. Expert Systems with Ap-
plications 41(17), 7980–7986 (2014)
22. And`o, B., Baglio, S., Lombardo, C.O., Marletta, V.: An
event polarized paradigm for adl detection in aal context.
IEEE Transactions on instrumentation and Measurement
64(7), 1814–1825 (2015)
23. Ozcan, K., Velipasalar, S.: Wearable camera-and
accelerometer-based fall detection on portable devices.
IEEE Embedded Systems Letters 8(1), 6–9 (2016)
24. Delahoz, Y.S., Labrador, M.A.: Survey on fall detection
and fall prevention using wearable and external sensors.
Sensors 14(10), 19806–19842 (2014)
25. Perry, J.T., Kellog, S., Vaidya, S.M., Youn, J.-H., Ali, H.,
Sharif, H.: Survey and evaluation of real-time fall detec-
tion approaches. In: High-Capacity Optical Networks and
Enabling Technologies (HONET), 2009 6th International
Symposium On, pp. 158–164 (2009). IEEE
26. Igual, R., Medrano, C., Plaza, I.: Challenges, issues and
trends in fall detection systems. Biomedical engineering
online 12(1), 66 (2013)
27. Wannenburg, J., Malekian, R.: Physical activity recogni-
tion from smartphone accelerometer data for user context
awareness sensing. IEEE Transactions on Systems, Man,
and Cybernetics: Systems (2016)
28. Ma, X., Wang, H., Xue, B., Zhou, M., Ji, B., Li, Y.:
Depth-based human fall detection via shape features
Fall detection monitoring systems: A comprehensive review 23
and improved extreme learning machine. IEEE journal
of biomedical and health informatics 18(6), 1915–1922
(2014)
29. Zhang, C., Lai, C.-F., Lai, Y.-H., Wu, Z.-W., Chao, H.-
C.: An inferential real-time falling posture reconstruction
for internet of healthcare things. Journal of Network and
Computer Applications (2017)
30. Hu, X., Qu, X.: An individual-specific fall detection
model based on the statistical process control chart.
Safety science 64, 13–21 (2014)
31. Naranjo-Hernandez, D., Roa, L.M., Reina-Tosina, J.,
Estudillo-Valderrama, M.A.: Personalization and adap-
tation to the medium and context in a fall detection
system. IEEE transactions on information technology in
biomedicine 16(2), 264–271 (2012)
32. Aslan, M., Sengur, A., Xiao, Y., Wang, H., Ince, M.C.,
Ma, X.: Shape feature encoding via fisher vector for ef-
ficient fall detection in depth-videos. Applied Soft Com-
puting 37, 1023–1028 (2015)
33. Kau, L.-J., Chen, C.-S.: A smart phone-based pocket fall
accident detection, positioning, and rescue system. IEEE
journal of biomedical and health informatics 19(1), 44–56
(2015)
34. Kwolek, B., Kepski, M.: Human fall detection on embed-
ded platform using depth maps and wireless accelerom-
eter. Computer methods and programs in biomedicine
117(3), 489–501 (2014)
35. Popescu, M., Hotrabhavananda, B., Moore, M., Skubic,
M.: Vampir-an automatic fall detection system using a
vertical pir sensor array. In: Pervasive Computing Tech-
nologies for Healthcare (PervasiveHealth), 2012 6th In-
ternational Conference On, pp. 163–166 (2012). IEEE
36. Yang, K., Ahn, C.R., Vuran, M.C., Aria, S.S.: Semi-
supervised near-miss fall detection for ironworkers with a
wearable inertial measurement unit. Automation in Con-
struction 68, 194–202 (2016)
37. Medrano, C., Plaza, I., Igual, R., S´anchez, ´
A., Castro,
M.: The effect of personalization on smartphone-based
fall detectors. Sensors 16(1), 117 (2016)
38. Yu, M., Yu, Y., Rhuma, A., Naqvi, S.M.R., Wang, L.,
Chambers, J.A.: An online one class support vector
machine-based person-specific fall detection system for
monitoring an elderly individual in a room environment.
IEEE journal of biomedical and health informatics 17(6),
1002–1014 (2013)
39. Li, Y., Ho, K., Popescu, M.: A microphone array sys-
tem for automatic fall detection. IEEE Transactions on
Biomedical Engineering 59(5), 1291–1301 (2012)
40. Debard, G., Karsmakers, P., Deschodt, M., Vlaeyen, E.,
Dejaeger, E., Milisen, K., Goedem´e, T., Vanrumste, B.,
Tuytelaars, T.: Camera-based fall detection on real world
data. Outdoor and large-scale real-world scene analysis,
356–375 (2012)
41. Liu, L., Popescu, M., Skubic, M., Rantz, M.: An au-
tomatic fall detection framework using data fusion of
doppler radar and motion sensor network. In: Engineer-
ing in Medicine and Biology Society (EMBC), 2014 36th
Annual International Conference of the IEEE, pp. 5940–
5943 (2014). IEEE
42. Medrano, C., Igual, R., Plaza, I., Castro, M., Fardoun,
H.M.: Personalizable smartphone application for detect-
ing falls. In: Biomedical and Health Informatics (BHI),
2014 IEEE-EMBS International Conference On, pp. 169–
172 (2014). IEEE
43. Kwolek, B., Kepski, M.: Improving fall detection by the
use of depth sensor and accelerometer. Neurocomputing
168, 637–645 (2015)
44. Kwolek, B., Kepski, M.: Fuzzy inference-based fall detec-
tion using kinect and body-worn accelerometer. Applied
Soft Computing 40, 305–318 (2016)
45. Wang, J., Zhang, Z., Li, B., Lee, S., Sherratt, R.: An
enhanced fall detection system for elderly person moni-
toring using consumer home networks. IEEE transactions
on consumer electronics 60(1), 23–29 (2014)
46. Rougier, C., Auvinet, E., Rousseau, J., Mignotte, M., Me-
unier, J.: Fall detection from depth map video sequences.
In: International Conference on Smart Homes and Health
Telematics, pp. 121–128 (2011). Springer
47. Hakim, A., Huq, M.S., Shanta, S., Ibrahim, B.: Smart-
phone based data mining for fall detection: analysis and
design. Procedia computer science 105, 46–51 (2017)
48. Shen, V.R., Lai, H.-Y., Lai, A.-F.: The implementation
of a smartphone-based fall detection system using a high-
level fuzzy petri net. Applied Soft Computing 26, 390–
400 (2015)
49. Hu, X., Qu, X.: Detecting falls using a fall indicator
defined by a linear combination of kinematic measures.
Safety science 72, 315–318 (2015)
50. Huang, S., Yang, Y., Liu, W.: An enhanced fall detection
approach based on cost sensitivity analysis. In: Software
and Network Engineering (SSNE), 2011 First ACIS In-
ternational Symposium On, pp. 81–85 (2011). IEEE
51. Kangas, M., Konttila, A., Lindgren, P., Winblad, I.,
J¨ams¨a, T.: Comparison of low-complexity fall detection
algorithms for body attached accelerometers. Gait & pos-
ture 28(2), 285–291 (2008)
52. Steidl, S., Schneider, C., Hufnagl, M.: Fall Detection by
Recognizing Patterns in Direction Changes of Constrain-
ing Forces. na, ??? (2012)
53. Wu, W., Dasgupta, S., Ramirez, E.E., Peterson, C., Nor-
man, G.J.: Classification accuracies of physical activities
using smartphone motion sensors. Journal of medical In-
ternet research 14(5), 130 (2012)
54. And`o, B., Baglio, S., Lombardo, C.O., Marletta, V.: A
multisensor data-fusion approach for adl and fall classifi-
cation. IEEE Transactions on Instrumentation and Mea-
surement 65(9), 1960–1967 (2016)
55. Colon, L.N.V., DeLaHoz, Y., Labrador, M.: Human fall
detection with smartphones. In: Communications (LAT-
INCOM), 2014 IEEE Latin-America Conference On, pp.
1–7 (2014). IEEE
56. Ghasemzadeh, H., Jafari, R., Prabhakaran, B.: A body
sensor network with electromyogram and inertial sensors:
multimodal interpretation of muscular activities. IEEE
transactions on information technology in biomedicine
14(2), 198–206 (2010)
57. Nguyen, T.-T., Cho, M.-C., Lee, T.-S.: Automatic fall
detection using wearable biomedical signal measurement
terminal. In: Engineering in Medicine and Biology Soci-
ety, 2009. EMBC 2009. Annual International Conference
of the IEEE, pp. 5203–5206 (2009). IEEE
58. Doukas, C., Maglogiannis, I.: Advanced patient or elder
fall detection based on movement and sound data. In:
Pervasive Computing Technologies for Healthcare, 2008.
PervasiveHealth 2008. Second International Conference
On, pp. 103–107 (2008). IEEE
59. Leone, A., Rescio, G., Caroppo, A., Siciliano, P.: A wear-
able emg-based system pre-fall detector. Procedia Engi-
neering 120, 455–458 (2015)
60. Karantonis, D.M., Narayanan, M.R., Mathie, M., Lovell,
N.H., Celler, B.G.: Implementation of a real-time human
movement classifier using a triaxial accelerometer for am-
bulatory monitoring. IEEE transactions on information
technology in biomedicine 10(1), 156–167 (2006)
24 Pranesh Vallabh, Reza Malekian
61. Zhang, T., Wang, J., Xu, L., Liu, P.: Fall detection by
wearable sensor and one-class svm algorithm. Intelligent
computing in signal processing and pattern recognition,
858–863 (2006)
62. Bourke, A.K., Lyons, G.M.: A threshold-based fall-
detection algorithm using a bi-axial gyroscope sensor.
Medical engineering & physics 30(1), 84–90 (2008)
63. Anania, G., Tognetti, A., Carbonaro, N., Tesconi, M.,
Cutolo, F., Zupone, G., De Rossi, D.: Development of a
novel algorithm for human fall detection using wearable
sensors. In: Sensors, 2008 IEEE, pp. 1336–1339 (2008).
IEEE
64. Lai, C.-F., Chang, S.-Y., Chao, H.-C., Huang, Y.-M.: De-
tection of cognitive injured body region using multiple
triaxial accelerometers for elderly falling. IEEE Sensors
Journal 11(3), 763–770 (2011)
65. Bianchi, F., Redmond, S.J., Narayanan, M.R., Cerutti,
S., Lovell, N.H.: Barometric pressure and triaxial
accelerometry-based falls event detection. IEEE Transac-
tions on Neural Systems and Rehabilitation Engineering
18(6), 619–627 (2010)
66. Yang, S.-W., Lin, S.-K.: Fall detection for multiple pedes-
trians using depth image processing technique. Computer
methods and programs in biomedicine 114(2), 172–182
(2014)
67. Stone, E.E., Skubic, M.: Fall detection in homes of
older adults using the microsoft kinect. IEEE journal of
biomedical and health informatics 19(1), 290–301 (2015)
68. Litvak, D., Zigel, Y., Gannot, I.: Fall detection of elderly
through floor vibrations and sound. In: Engineering in
Medicine and Biology Society, 2008. EMBS 2008. 30th
Annual International Conference of the IEEE, pp. 4632–
4635 (2008). IEEE
69. Sposaro, F., Tyson, G.: ifall: an android application for
fall monitoring and response. In: Engineering in Medicine
and Biology Society, 2009. EMBC 2009. Annual Inter-
national Conference of the IEEE, pp. 6119–6122 (2009).
IEEE
70. Alwan, M., Rajendran, P.J., Kell, S., Mack, D., Dalal, S.,
Wolfe, M., Felder, R.: A smart and passive floor-vibration
based fall detector for elderly. In: Information and Com-
munication Technologies, 2006. ICTTA’06. 2nd, vol. 1,
pp. 1003–1007 (2006). IEEE
71. Werner, F., Diermaier, J., Schmid, S., Panek, P.: Fall
detection with distributed floor-mounted accelerometers:
An overview of the development and evaluation of a fall
detection system within the project ehome. In: Perva-
sive Computing Technologies for Healthcare (Pervasive-
Health), 2011 5th International Conference On, pp. 354–
361 (2011). IEEE
72. Popescu, M., Mahnot, A.: Acoustic fall detection using
one-class classifiers. In: Engineering in Medicine and Bi-
ology Society, 2009. EMBC 2009. Annual International
Conference of the IEEE, pp. 3505–3508 (2009). IEEE
73. Chaccour, K., Darazi, R., el Hassans, A.H., Andres, E.:
Smart carpet using differential piezoresistive pressure
sensors for elderly fall detection. In: Wireless and Mobile
Computing, Networking and Communications (WiMob),
2015 IEEE 11th International Conference On, pp. 225–
229 (2015). IEEE
74. Yazar, A., Keskin, F., T¨oreyin, B.U., C¸etin, A.E.: Fall de-
tection using single-tree complex wavelet transform. Pat-
tern Recognition Letters 34(15), 1945–1952 (2013)
75. Tomii, S., Ohtsuki, T.: Falling detection using multiple
doppler sensors. In: e-Health Networking, Applications
and Services (Healthcom), 2012 IEEE 14th International
Conference On, pp. 196–201 (2012). IEEE
76. Rimminen, H., Lindstr¨om, J., Linnavuo, M., Sepponen,
R.: Detection of falls among the elderly by a floor sensor
using the electric near field. IEEE Transactions on In-
formation Technology in Biomedicine 14(6), 1475–1476
(2010)
77. Zhuang, X., Huang, J., Potamianos, G., Hasegawa-
Johnson, M.: Acoustic fall detection using gaussian mix-
ture models and gmm supervectors. In: Acoustics, Speech
and Signal Processing, 2009. ICASSP 2009. IEEE Inter-
national Conference On, pp. 69–72 (2009). IEEE
78. Nizam, Y., Mohd, M.N.H., Jamil, M.M.A.: Human fall
detection from depth images using position and veloc-
ity of subject. Procedia Computer Science 105, 131–137
(2017)
79. Miaou, S.-G., Sung, P.-H., Huang, C.-Y.: A customized
human fall detection system using omni-camera images
and personal information. In: Distributed Diagnosis and
Home Healthcare, 2006. D2H2. 1st Transdisciplinary
Conference On, pp. 39–42 (2006). IEEE
80. Krekovi´c, M., ˇ
Ceri´c, P., Dominko, T., Ilijaˇs, M., Ivanˇci´c,
K., Skolan, V., ˇ
Sarlija, J.: A method for real-time detec-
tion of human fall from video. In: MIPRO, 2012 Proceed-
ings of the 35th International Convention, pp. 1709–1712
(2012). IEEE
81. Thome, N., Miguet, S.: A hhmm-based approach for ro-
bust fall detection. In: Control, Automation, Robotics
and Vision, 2006. ICARCV’06. 9th International Confer-
ence On, pp. 1–8 (2006). IEEE
82. Rougier, C., Meunier, J., St-Arnaud, A., Rousseau, J.:
Fall detection from human shape and motion history us-
ing video surveillance. In: Advanced Information Net-
working and Applications Workshops, 2007, AINAW’07.
21st International Conference On, vol. 2, pp. 875–880
(2007). IEEE
83. Yun, Y., Gu, I.Y.-H.: Human fall detection in videos by
fusing statistical features of shape and motion dynamics
on riemannian manifolds. Neurocomputing 207, 726–734
(2016)
84. Foroughi, H., Rezvanian, A., Paziraee, A.: Robust fall de-
tection using human shape and multi-class support vector
machine. In: Computer Vision, Graphics & Image Pro-
cessing, 2008. ICVGIP’08. Sixth Indian Conference On,
pp. 413–420 (2008). IEEE
85. Hazelhoff, L., Han, J., et al.: Video-based fall detection in
the home using principal component analysis. In: Interna-
tional Conference on Advanced Concepts for Intelligent
Vision Systems, pp. 298–309 (2008). Springer
86. Auvinet, E., Multon, F., Saint-Arnaud, A., Rousseau, J.,
Meunier, J.: Fall detection with multiple cameras: An
occlusion-resistant method based on 3-d silhouette verti-
cal distribution. IEEE Transactions on Information Tech-
nology in Biomedicine 15(2), 290–300 (2011)
87. Rougier, C., Meunier, J., St-Arnaud, A., Rousseau, J.:
Monocular 3d head tracking to detect falls of elderly
people. In: Engineering in Medicine and Biology Society,
2006. EMBS’06. 28th Annual International Conference of
the IEEE, pp. 6384–6387 (2006). IEEE
88. Hsu, Y.-t., Hsieh, J.-w., Kao, H.-F., Liao, H.-y.M.: Hu-
man behavior analysis using deformable triangulations.
In: Multimedia Signal Processing, 2005 IEEE 7th Work-
shop On, pp. 1–4 (2005). IEEE
89. De Maio, C., Fenza, G., Loia, V., Orciuoli, F.: Making
sense of cloud-sensor data streams via fuzzy cognitive
maps and temporal fuzzy concept analysis. Neurocom-
puting (2017)
90. Cao, Y., Yang, Y., Liu, W.: E-falld: A fall detection
system using android-based smartphone. In: Fuzzy Sys-
Fall detection monitoring systems: A comprehensive review 25
tems and Knowledge Discovery (FSKD), 2012 9th Inter-
national Conference On, pp. 1509–1513 (2012). IEEE
91. Chen, O.T.-C., Kuo, C.-J.: Self-adaptive fall-detection
apparatus embedded in glasses. In: Engineering in
Medicine and Biology Society (EMBC), 2014 36th An-
nual International Conference of the IEEE, pp. 4623–
4626 (2014). IEEE
92. Schwickert, L., Becker, C., Lindemann, U., Mar´echal,
C., Bourke, A., Chiari, L., Helbostad, J., Zijlstra, W.,
Aminian, K., Todd, C., et al.: Fall detection with body-
worn sensors. Zeitschrift f¨ur Gerontologie und Geriatrie
46(8), 706–719 (2013)
A preview of this full-text is provided by Springer Nature.
Content available from Journal of Ambient Intelligence and Humanized Computing
This content is subject to copyright. Terms and conditions apply.
