International Journal of Integrated Engineering, Vol. 8 No. 1 (2016) p.

35-43

A Study on Human Fall Detection Systems: Daily Activity

Classification and Sensing Techniques
Yoosuf Nizam1,*, Mohd Norzali Haji Mohd1,2, M. Mahadi Abdul Jamil1
Yoosuf Nizam1,*, Mohd Norzali Haji Mohd2, M. Mahadi Abdul Jam
1
Biomedical Engineering Modeling and Simulation (BIOMEMS) Research Group,
2
Embedded Computing Systems (EmbCos),
Faculty of Electrical and Electronic Engineering,
Universiti Tun Hussein Onn Malaysia, 86400 Johor, MALAYSIA.

Abstract: Fall detection for elderly is a major topic as far as assistive technologies are concerned. This is due to
the high demand for the products and technologies related to fall detection with the ageing population around the
globe. This paper gives a review of previous works on human fall detection devices and a preliminary results from
a developing depth sensor based device. The three main approaches used in fall detection devices such as wearable
based devices, ambient based devices and vision based devices are identified along with the sensors employed.
The frameworks and algorithms applied in each of the approaches and their uniqueness is also illustrated. After
studying the performance and the shortcoming of the available systems a future solution using depth sensor is also
proposed with preliminary results.
Keywords: Fall Detection, Algorithm, Approach, Depth Sensor, Assistive technology

1. Introduction and enhance the health care services too. There have been
plenty of researches conducted in this area to develop
Assistive technology or adaptive technology is an
systems and algorithms for enhancing the functional
emerging topic since daily living assistance are very often
ability of the elderly people patients with special needs
needed for many people in today’s aging populations
such as disable [18]. This has led to the improvement in
including disabled, over weight, obese and elderly people.
the technologies used to make such systems and thus
One important aim of assistive technology is to provide
make it feasible. The confidential level of the devices has
better health care to those in need, especially allowing
also increased which encouraged people to use the
elderly people who live alone to stay independently as
product. This in turn lead to the reduction in labor cost in
long as possible in their own home without changing their
terms of no presence of medical staffs at all times looking
life style.
after the elderly people.
In order to provide better living for them, it is
The recent researches conducted on assistive health
important to have continues human monitoring systems in
monitoring technologies for elderly people can be
their home to inform the health care representatives of
categorized into five classes [18-20]. These classes
any emergency attendance. Among such monitoring
distinguish different detection methods used, as wearable
systems, fall detection systems are in increasing interest
device based, wireless based, ambiance based, vision
since statistics [1, 2] shows that falls are the main reasons
based and floor sensor/electric field sensor based.
of injury related death for seniors aged 79 [3, 4] or above
This paper presents a short review of the studies on
and it is the second common cause of injury related
fall detection systems, which could serve as a reference to
(unintentional) death for all ages [5, 6]. Also fall is the
the researchers in the area for further investigation. Fall
biggest threat among all other incidents to elderly and
types and different fall detection approaches are
those people in need [3, 7-16]. For elderly people, fall can
reviewed, along with the algorithms used to detect fall
have severe consequences, especially if not attended in a
and fall detection sensors. Finally, some initial results
short period of time [17]. Accurate independent fall
from a depth sensor based approach with Microsoft
detection systems are very important to help the elderly
Kinect v1 sensor is presented, which requires further
people to live independently because it has proved that
improvement to develop a reliable fall detection system.
the medical consequences of a fall are highly dependent
on the response and rescue time of the medical staff [18],
which is only possible with accurate and reliable fall 2. Characteristics of Fall
detection mechanism. Such detections are also vital since Falls can be categorized into different types
there may be a case where someone lose consciousness or depending on the characteristics of the movements that
are unable to call for help. lead to the fall. It is important to recognize these
Therefore highly accurate fall detection can characteristics of the movement in order to understand the
significantly improve the living of lonely elderly people

*Corresponding author: [email protected] 35

2016 UTHM Publisher. All right reserved.
penerbit.uthm.edu.my/ojs/index.php/ijie
Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

existing algorithms used to detect falls and also to device Fig. 1: Hierarchy of fall detection methods.
new algorithms.
Falls can be divided into many types from the As illustrated in fig 1, wearable device based systems
characteristics of movements that causes falls. Yu [20] can be further divided into two different categories based
divided the falls into four types. Fall from sleeping (bed), on the approach used, as inactivity based (motion based)
fall from sitting (chair), fall from walking or standing on and posture based. Similarly ambient/fusion based
the floor and fall from standing on support such as ladder, devices either use floor sensors or electric field and
tools [20]. The characteristics of the four types of falls are posture based sensors to detect fall. Three different
significantly different from each other even though they approaches are used in camera or vision based devices.
do share some crucial features [18]. The last type of fall is These different approaches and methods of fall
not common amongst elderly people since they normally detection as mentioned above do share same general
occur among working people. The main challenge that framework as illustrated in fig 2a [20]. Some of the fall
the researchers are facing is some of the characteristics detection methods just use one sensor indicator with a
mentioned above do exists in normal daily actions such as threshold while others use complicated algorithms and
crouch. image processing to detect falls. The only distinguishing
factor among them is the sensors deployed, the number of
3. Classification of Fall Detection sensors used and their detection algorithms. For example,
Approaches data acquisition for sensing one indicator can be different
from a single sensor than to multiple sensors and different
Human fall detection system classifies activities of cameras working together to collect data [20]. The
daily life to identify an unintentional fall. Various sensors framework followed in three main types of fall detection
and techniques are been used to classify daily activities to systems also differ in architectural design and the
distinguish human fall. Noury et al. reviewed the communication methods implemented between the inner
algorithms and sensors for automatic early fall detection components. Fig 2b, fig 2c and fig 2d illustrate the
for elderly people [21]. In 2008 Yu also gave a survey on general framework for the three main approaches of fall
fall detection for elderly and patients focusing on the detection as wearable, ambient and camera based
approaches and principles of fall detection [20]. Mubashir approaches respectively. These different approaches of
et al. also wrote a comprehensive survey on recent fall fall detection are briefly discussed in the next section
detection techniques [18]. These are the only review with some review of the previous works in the following
papers on principles and methods used in fall detection. sub-sections.
Therefore this paper is meant to briefly review the scopes

communication
and challenges of the fall detection principles and Data Fall alert
Acquisition

Sensors
from input

Wired or
approaches and then propose a future solution that can be processing decision Distance

wireless
Data

implemented to develop a new system framework that and image and care
Camera
processing generation givers site
would be widely accepted by the elderly people, patients
and care givers. (a)
There are various methods that can be used to detect communication
human fall such as using a camera to sense a fall or using Data processing,
Acquisition
from input

Wired or

various sensors to detect fall. These sensors use different

wireless

Distance
Data

Sensors fall alert decision

approaches to identify a fall. In this respect, from the care
analysis of fall detection methods they can be divided Bluetooth, and generation givers site
into three main approaches which could be further Rf etc
(b)
divided into few other categories depending on the
communication

sensors and algorithms used [22]. The three basic

Wired or

approaches are wearable device based, camera based

wireless

Sensors Data Data processing, Distance

devices and ambience sensor based devices [18, 20, 22, Acquisition fall alert decision care
23] as shown in fig 1. from input and generation givers site
(c)
Fall detection
communication
Acquisition

Camera
from input

Image processing, Distance

Wired or
wireless

(Video and
Data

Wearable device Camera / Ambience / fall alert decision care

depth
based vision based fusion based and generation givers site
images)

Inactivity Posture Floor Sensor / (d)

Posture
based based Electric field
Fig. 2: Framework for different approaches used in fall
detection (a) General framework used in all fall detection
Body shape change Inactivity 3D head change and alerting (b) Framework for wearable based
Analysis detection Analysis

36
 Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

approaches (c) Framework for ambient based approaches interpretive method for characterizing acceleration and
(d) Framework for Camera based approaches. body position. It is remote monitoring approach for
detecting serious fall events in nursing home patients or
3.1 Wearable Based Technique in similar care facilities.
Bianchi et al used barometric pressure sensor, as a
Most of the previous academic researches on fall
surrogate measure of altitude to help in differentiating
detection were based on wearing devices with embedded
real fall events from normal daily living activities. The
sensors to detect posture and motion of the body. Those
waist based device measures the acceleration and air
sensors are either placed on garments or they are in the
pressure data which is then analyzed offline [32].
form of wearable belts. Different types of sensors are
Estudillo-Valderrama et al used a Personal Server for
used to make such devices and hence there are varieties
controlling and processing of data acquired from multiple
of methods to detect falls, which are briefly summarized
biomedical sensors to detect human fall. They focused on
below.
PSE (Personal server) which is an electronic device that
can be worn as a watch or a pendent. Future
3.1.1 Accelerometer Based Devices improvements include optimization of communication
Among the wearable devices, accelerometer is the protocols between PSE and smart accelerometer sensors
most extensively used method to realize a fall. It uses the and a second validation stage of the proposed algorithms
measure of the acceleration of the body to classify falls. with actual elderly people [33].
Clifford et al patented a human body fall detection Tamura et al developed a wearable airbag that
system using accelerometer, a processor and a wireless incorporates in a fall detection system that uses both
transmitter [24]. The processor uses accelerometer acceleration and angular velocity signals to trigger
measurements to determine if the person wearing the inflation of the airbag. Apart from the wearing difficulties
device is falling and also a non-movement phase followed and false alarm can give discomfort to the wearer as the
by a fall. The generated response is then transmitted trigger will inflate the air bag [34].
remotely to a signal receiver by the wireless transmitter Chen et al created a wireless low power sensor
[24]. network by using small, non-invasive, low power motes
Doukas et al. developed an accelerometer based (sensor nodes). The sampling of acceleration is done
wearable device to detect human falls for patients. Along sequentially on-board device to reduce the burden onto
with the accelerometer data, Support Vector Machine the network. For fall detection the angle of change is
(SVM) is used to determine fall. Data is collected produced by the dot product of acceleration vector, from
wirelessly from an accelerometer which is attached at the the orientation information [35].
foot of the user. [25]. Wang et al developed a system for fall detection that
Noury et al developed a device that collects vertical uses an accelerometer which is placed on the head. By
acceleration shock from a piezo electric accelerator, the using a reference velocity which is calculated using the
body orientation from a position tilt switch and the backward integration of accelerations they are able to
mechanical vibration of the body surface, which are then distinguish falls from normal daily activities with a
sent to a computer to detect human fall incident [26, 27]. predefined threshold [36].
Mathie et al used a waist mount accelerometer to
detect falls. Fall is detected when a negative acceleration 3.1.2 Posture Sensor Based Devices
is sensed from the change in orientation from upright to a
There are studies conducted on determining human
lying position of the waist mount accelerometer [28].
fall using the posture movements. Body orientation as
Nyan et al developed a garment based fall detector
posture is used to detect fall using either posture sensors
and daily activity monitor using 3 axis Micro Electric
or multiple accelerometers [37, 39].
Mechanical System (MEMS) accelerators to collect data.
Kang et al developed a wrist worn fall detector using
A signal processor using discrete wavelet transformation
two axial accelerometers with a posture sensor. It
(DWT), processes the collected data and identifies fall
includes a tele-reporting capability for emergency
and activity. Sensor is placed in shoulder position of a
telemedicine and telecare. Fall detection was part of the
jacket and all other components such as microcontroller, device which also includes single-channel
batteries and Bluetooth transmitter are inside the pocket. electrocardiogram (ECG), noninvasive blood pressure
Therefore it is subject to wearing difficulties and it also (NIBP), pulse oximetry (SpO2), respiration rate and body
not possible to wear it all the time. [29]. surface temperature measuring unit. Since it is wrist
Narayanan et al described a distributed fall
worn, the device’ accuracy in fall detection is subject to
management system capable of real time fall detection in
the different habits of hand movement of people [37].
an unsupervised living context using a waist mounted tri-
Ghasemzadeh et al created a physiological
axial accelerometer. The wearable unit has to be attached
monitoring system that collects acceleration and muscle
to the waist belt every morning and place back for
activity and performs analysis on those signals during
charging before they go to bed [30].
standing balance to assess the behavior of the
An elderly fall monitoring method and device that
electromyogram (EMG) signal to interpret the activity of
uses accelerators to obtain motion data was patented by
postural control system in terms of balance control [38].
Petelenz et al. in [31]. It uses signal feature extraction and

37
Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

Kaluza et al presented a posture based fall detection motions of daily activities [20-21]. More ever it is often
algorithm using the ideology of reconstruction of an rejected by the elderly because of the difficulty of the
object’s posture. The posture is reconstructed in a 3D wearing devices or garments. Irrespective of this, it do
plane by locating the wireless tags which were placed on have the advantages of been cheap, easy to setup and
body parts (sewn on clothes) such as hips, ankles knees, operate.
wrists elbows and shoulders. Some tags are also placed at
specific positions such as bed, chair, sofa, table to 3.2 Vision Based Technique
identify some postures such as lying on bed or sitting on
Vision or camera based devices are increasingly in
chair. The fall detection algorithms use acceleration
use due to its multiple advantages over other sensor based
thresholds along with velocity profiles. Acceleration is
devices [44-47]. Some of the reasons are that, cameras
derived from the movements of the tags. Acceleration and
can be used to detect multiple events simultaneously and
accurate velocity calculation is subject to the tag’s
it can avoid the difficulty of wearing devices and
localization precision [39].
garments for fall detection. Most importantly the recorded
Kangas et al used a waist worn tri-axial
video from camera can be used for verification after a fall
accelerometer, transceiver and microcontroller to develop
has occurred. Vision based approach do have
a new fall detector prototype based on fall associated
disadvantages of not preserving the users’ privacy.
impact and end posture [40].
Selected previous works on camera or vision based
devices from different sub-categories are briefly reviewed
3.1.3 Accelerometer and Gyroscope Based in this section.
Devices
Combination of accelerometer and gyroscope data 3.2.1 Inactivity
has also proved to determine fall more accurately than With this approach, a fall is detected based on the
any of the sensor alone. Accelerometer can provide inactivity period on the floor. Camera or motion detector
kinetic force while gyroscope can help to estimate the tracks the person to obtain motion traces and based on it a
current posture [41]. The combination of two sensors can fall is determined [48].
also help to identify any false measurement from anyone Jansen and Deklerck used a stereo camera to acquire
sensor. Some studies are presented below. depth images (3D images) to identify body area and to
Nyan et al used 3D accelerometer and 2D gyroscope find body orientation. The orientation change of the body
worn on thigh which is based on Body Area Network is used to detect inactivity and if inactivity occurs in
(BAN) to prevent from fall related injuries such as certain context a fall is detected [48].
inflatable airbag for hip protection before the impact. The
system is based on the concept that thigh segments will
not exceed a certain threshold angle to side and forward
3.2.2 Shape Change
in normal daily activities except for fall, which was The main perception with this approach is that the
validated in an experiment with 21 young healthy shape of a person will change from standing to lying if a
volunteers performing normal daily activities and falls fall occurs.
[41]. Toreyin et al presented a Hidden Markov Model
Li et al presented a novel fall detection system using (HMM) based fall detection where HMM uses video
both accelerometer and gyroscopes. By using two tri- features to differentiate fall from walking. And another
axial accelerometer at separate body locations they are HMM based approach uses audio features to differentiate
able to recognize four kinds of static postures: standing, falling sound from talking [49].
bending, sitting and lying. Motions between these static Another HMM-based algorithm to detect fall was
postures are considered as dynamic transitions and if the proposed by Anderson et al [50]. The HMM uses multiple
transition before lying posture is not intentional, a fall is features extracted from silhouette: height of bounding
detected. Whether motion transitions are intentional or box, magnitude of motion vector, determinant of
not is determined by the linear acceleration and angular covariance matrix and ratio of width to height of
velocity measurements [42]. bounding box of person [50].
Bourke et al used a bi-axial gyroscope sensor Thome and Miguet proposed robust Hierarchical
mounted on the trunk to differentiate fall from normal Hidden Markov Model (HHMM) based algorithm to
daily activities. Fall is determined from measurement of detect fall, where HHMM is used to model the motion.
pitch and roll angular velocities and a threshold-based Many improvements are possible including automating
algorithm [43]. the rectification processes using Hough transfer for
detecting sets of parallel lines and computing the
3.1.4 Discussion on Wearable Based orthogonal vanishing points. Optimal Placement of
additional cameras for further improved recognition can
Approach be achieved by using the relationship between image
In all of the wearable based devices discussed above, angle and posture [51].
a fall is distinguished from normal daily activities using Miaou et al used a rule based algorithm with an
the unique pattern of motion the fall possess. Therefore it Omni-camera and uses context information for fall
is prone to give false alarm by triggering a fall from

38
 Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

detection. Fall is determined based on the ratio of width significantly different from normal daily activities and at
to height of the bounding box of body in image [52, 53]. the same it is also different from the vibration generated
A 3D shape of body extracted from multiple cameras by objects falling on the floor [56].
to detect fall was proposed by Cucchira et al. This multi- Scott patented a bed exit detection apparatus by
camera vision system for detecting and tracking people using bladder or other fluid carrying devices in fluid
employs warping people’s silhouette technique to communication with a pressure sensor. This particularly a
exchange visual information between overlapped cameras patient presence detection system that enables the care
when camera handover occurs. The system also delivers giver to get alert about both the presence and absence of
live video streams to operators/remote users in order to the patient on patient carrying surface. Especially,
verify the received alarm [54]. whether the patient is sitting upright or is leaning, falling
forward or falling sideways [57].
3.2.3 3D Head Motion Analysis Sixsmith et al used an array of pyroelectric IR
sensors on wall for detecting fall. The infrared arrays will
The principle used in this approach is that during a
locate and track the thermal target in the sensor’s field of
human fall, vertical motion is faster than the horizontal
view. The array also provides size, location and velocity
motion [55].
information. They considered the motion of target and
Rouger et al developed fall detector using monocular
inactivity period to identify fall. Further improvements
3D head tracking. They separated fall from walking by
include improving the fall detection algorithm, creating
computing the vertical and horizontal velocity of head
algorithm to track, locate multiple individuals [58, 59].
with two threshold values. The method presented is based
A wireless sensor network with array of sensors and
on Motion History Image (MHI) and changes in human
event detection and modalities and distributed processing
shape. The detection method is based on the fact that the
for smart home monitoring application was proposed by
motion is large when a fall occurs. Therefore the system
Tabar et al [60].
will first detect a large motion and if a motion is detected
Zhuang et al proposed a system that detects human
than the shape of the person in the video sequence will be
falls in the home environment, distinguishing them from
analyzed. In this stage the concept used is that during a
competing noise, by using audio signal from a single far-
fall the human shape changes and at the end of the fall the
field microphone. They model each fall or noise segment
person is usually on the floor with few second and with
using a GMM super vector to distinguish them from
less body movements [55].
background noise and classify the audio segments into
falls and other types of noise using SVM built on GMM
3.2.4 Discussion on Vision Based Approach super vector kernel [61].
This is the most reliable technique for fall detection
compared to the other approaches [53-55]. If individual 3.3.2 Discussion on Ambient Based Approach
different sub-categories are compared, inactivity
This approach also has several disadvantages like
detection is simple in terms of processing and hence they
wearable devices. Since it is mostly based on pressure
are less reliable. A shape detection algorithm is more
sensor which is very prone to measure weight of all
reliable because body shape detection can give more objects, thus false alarm rate is very high. Unlike
accurate information about fall than head detection. And wearable devices, it causes fewer disturbances to the
3D body shape detection uses more cameras and requires users. Similar to wearable devices it is very cost effective
complex computing. The recent trend in fall detection
and does not require high installation costs [56, 59].
uses depth sensor for human identification and movement
recognition. The three techniques discussed in vision
based approach namely inactivity detection, shape change 4. Results from Depth Sensor
and 3D head motion analysis can be applied with a single For the experimental setup, Microsoft® Kinect v1
depth sensor. The section 4 shows some preliminary Sensor was used as a depth sensor to generate the
result of fall detection system based on depth sensor necessary depth images. Different activities of daily life
approach. such as walking, running, sitting on chair, sitting on floor,
lying of floor, fall from chair, fall while standing and fall
3.3 Ambient Based Technique while trying to sit on a chair have been tested on the
experimental setup to obtain results for performance
This approach usually uses array of sensors to
analysis of the system. Algorithms employed used,
identify falls through pressure sensing, vibrational data,
changes of subject’s height and body velocity for the
IR sensor and single far-field microphone [56, 58-61].
classification of human fall from other activities of daily
life. The results obtained and the methodologies that can
3.3.1 Pressure Sensing and Vibration be applied after studying the performance of the proposed
Alwan et al used array of vibration sensors on the system is further discussed in the following paragraphs.
floor and a processor to identify fall by analyzing location From the review it was found that a better fall
data. The methods used is based on the perception that a detection could be developed using depth sensors rather
human fall will always cause a vibration pattern on the than conventional color camera or wearable and other
floor and implies that the vibration generated from fall is non-invasive sensors. Since this approach is also non-

39
Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

invasive and are far better in human movement detection changing pattern can be used together with the velocity
using 3D depth maps. This section will introduce some and position of joints to classify human fall from other
methods of measurements that can be implemented with a activities of daily life. As seen from fig. 3, the changes of
depth sensor to classify human fall from other activities height pattern itself can distinguish some of the activities
of daily life. like sitting on chair and sitting of floor. For those
One method to identify a fall would be to measure activities where the height change patterns are similar and
the human height continuously. Human fall can be are difficult to distinguish, the changes of velocity has
predicted, if the height drops close to the floor plane shown clear gaps as illustrated in fig. 4, which shows the
either from a height which can be assumed as standing or changes of head position for falling and lying on floor
sitting on a chair. On the other hand lying down on floor from standing. Experimental results also showed that
will also give a similar drop of human height. Therefore these two movements are most difficult to distinguish
the velocity or acceleration of the body has to be since the changes of height pattern are very similar except
incorporated to accurately distinguish lying down on the the time it takes. Even in this stage the system showed
floor from a fall. Apart from these measurements, the good performance in classifying fall from other activities
position of human subject could also provide information except intentional lying, this is an indication that further
in classifying similar activities and also identifying the improvements on the algorithm can increase the accuracy
posture. The pattern of human height changes for of the system. Thus further study can lead to development
different activities is shown in Fig. 3. These heights of a better fall detection system using depth sensor.

Fig 3: Result of changes in distance for the activities (a) Standing; (b) bend and stand up; (c) Walking across the
sensor; (d) Walking slowly across the sensor; (e) Running; (f) Walking around the sensor; (g) sit down on floor; (h)
falling from standing; (i) sit on chair; (j) fall from chair; (k) stand up.

Fig 4: Changes of velocity for a fall from standing and lying on floor from standing.

It is obvious that the velocity for “falling” is higher and the changes are rapid than for lying on floor. These
changes can be used in conjunction with distance from head to floor to separate the two activities.

40
 Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

5. Summary and Future Works problems and provide a better solution. From this point of
view a better fall detection could be developed by
This paper has presented a review of different
incorporating accelerometers or gyroscope with camera
approaches used in human fall detection for elderly
based approaches to improve the detection accuracy.
people. The underlining features and methods along with
Since devices with only accelerometer or gyroscope have
the algorithms for detecting falls are also described. The
shown high false alarm rates, integrating a camera for fall
merits, reliability and pitfalls in the existing approaches
verification could solve the problem to some extent, but it
are briefly highlighted. Table 1 compares the existing
is again subject to privacy and confidentiality issues.
system approaches based on their performance, accuracy,
Since camera will be used for verification only, the issue
acceptably, reliability, recognition and cost. We have also
of privacy will be less of a concern. New techniques and
shown some results generated from preliminary testing
algorithms can also be developed to do verification within
conducted on a fall detection system based on depth
the system rather than transmitting images to caregivers
sensor. The results showed that the depth sensor can
for fall confirmation. Additional works on reducing the
classify human fall from other activities of daily life more
complexity of fall detection in vision based systems from
accurately and is reliable. At the same time it can
low light conditions and of course fall prevention systems
preserve the privacy of users, since color videos are not
can help to reduce fall occurrence. However the best
captured.
solution as per the preliminary results obtained is the
On the other hand wearable sensor based devices
depth sensor based approach. It would be a better than
sometimes cannot distinguish fall from sitting especially
many other approaches in terms of the parameters
if it is accelerometer based. Also it is very often rejected
described in table 1, except for the cost and setup. In
by the wearer due to the difficulty of wearing such belts
terms of the cost, a depth sensor based fall detection
or garments and discomforts associated with it. Ambient
system could be similar to a normal color camera based
based devices are also prone to generate false alarms
system and it could be cheaper if the color camera based
mainly from the pressure sensors which measure the
system incorporates multiple cameras. The setup
pressure of everything. Apart from these two methods the
difficulty would be same as a color camera based system
vision based approach has high reliability and accuracy. It
or it can depend on how the system is developed. From
also avoids the disturbance to the wearer that otherwise is
the review and results obtained it can be concluded that
caused by the wearing and other mounted devices. Vision
depth sensor can be used to develop an accurate non-
based devices can also provide verification of falls after
invasive fall detection system with least false alarm ratio.
the event has occurred. At the same time it has the
At the same time the problem of obstacles around can be
disadvantage of not preserving the user’s privacy which
minimized by using multiple sensors on different
could be solved by using depth sensors.
orientations.
Irrespective of the issues mentioned in the existing
methods and devices, a baseline has been established to
help developing new techniques which can minimize the

Table 1: Comparison of different approaches used in fall detection

Cost and Performance Properties
Approach Sub-category cost Recognition rate Setup Reliability acceptability Accuracy
Inactivity
Wearable Cheap low easy no Not always low
Posture
Floor sensors Cheaper than yes
Ambient Low to medium easy depends medium
Posture camera based Depends on users
Body shape
Medium Good Depends Good High
change
Inactivity Medium Good Medium Good Depends on users High
Vision based
3D head change Medium Good Medium Good High
Depth sensor Medium High Medium High High High

[2] Baker, S.P. and A. Harvey, Fall injuries in the

References elderly. Clinics in geriatric medicine, 1985. 1(3): p.
501-512.
[1] Griffiths, C., C. Rooney, and A. Brock, Leading [3] Stevens, J.A., et al., The costs of fatal and non-fatal
causes of death in England and Wales–how should falls among older adults. Injury prevention, 2006.
we group causes. Health Statistics Quarterly, 2005. 12(5): p. 290-295.
28(9). [4] Kannus, P., et al., Prevention of falls and consequent
injuries in elderly people. The Lancet, 2005.
366(9500): p. 1885-1893.

41
Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

[5] Bourke, A., J. O’brien, and G. Lyons, Evaluation of a Applications and Services, 2008. HealthCom 2008.
threshold-based tri-axial accelerometer fall detection 10th International Conference on. 2008. IEEE.
algorithm. Gait & posture, 2007. 26(2): p. 194-199. [21] Noury, N., et al. Fall detection-principles and
[6] Kangas, M., et al., Comparison of low-complexity methods. in Engineering in Medicine and Biology
fall detection algorithms for body attached Society, 2007. EMBS 2007. 29th Annual
accelerometers. Gait & posture, 2008. 28(2): p. 285- International Conference of the IEEE. 2007. IEEE.
291. [22] Luo, S. and Q. Hu. A dynamic motion pattern
[7] Gostynski, M., [Prevalence, circumstances and analysis approach to fall detection. in Biomedical
consequences of falls in institutionalized elderly; a Circuits and Systems, 2004 IEEE International
pilot study]. Sozial-und Praventivmedizin, 1990. Workshop on. 2004. IEEE.
36(6): p. 341-345. [23] Kepski, M., B. Kwolek, and I. Austvoll. Fuzzy
[8] Gurley, R.J., et al., Persons found in their homes inference-based reliable fall detection using Kinect
helpless or dead. New England Journal of Medicine, and accelerometer. in Artificial Intelligence and Soft
1996. 334(26): p. 1710-1716. Computing. 2012. Springer.
[9] Lin, L.-J., F.-T. Chiou, and H.H. Cohen, Slip and fall [24] Clifford, M.A., R.L. Borras, and L. Gomez, System
accident prevention: A review of research, practice, and method for human body fall detection. 2007,
and regulations. Journal of safety research, 1996. Google Patents.
26(4): p. 203-212. [25] Doukas, C., et al., Patient fall detection using support
[10] Lord, C.J. and D.P. Colvin. Falls in the elderly: vector machines, in Artificial Intelligence and
Detection and assessment. in Engineering in Innovations 2007: from Theory to Applications.
Medicine and Biology Society, 1991. Vol. 13: 1991., 2007, Springer. p. 147-156.
Proceedings of the Annual International Conference [26] Noury, N., et al. Monitoring behavior in home using
of the IEEE. 1991. IEEE. a smart fall sensor and position sensors. in
[11] Lord, S.R., et al., Falls in older people: risk factors Microtechnologies in Medicine and Biology, 1st
and strategies for prevention. 2007: Cambridge Annual International, Conference On. 2000. 2000.
University Press. IEEE.
[12] Sadigh, S., et al., Falls and fall-related injuries [27] Noury, N., et al. A smart sensor based on rules and
among the elderly: a survey of residential-care its evaluation in daily routines. in Engineering in
facilities in a Swedish municipality. Journal of medicine and biology society, 2003. Proceedings of
community health, 2004. 29(2): p. 129-140. the 25th annual international conference of the IEEE.
[13] Salvà, A., et al., Incidence and consequences of falls 2003. IEEE.
among elderly people living in the community. [28] Mathie, M.J., et al., Accelerometry: providing an
Medicina clinica, 2004. 122(5): p. 172-176. integrated, practical method for long-term,
[14] Teasell, R., et al., The incidence and consequences of ambulatory monitoring of human movement.
falls in stroke patients during inpatient rehabilitation: Physiological measurement, 2004. 25(2): p. R1.
factors associated with high risk. Archives of [29] Nyan, M., et al. Garment-based detection of falls and
physical medicine and rehabilitation, 2002. 83(3): p. activities of daily living using 3-axis MEMS
329-333. accelerometer. in Journal of Physics: Conference
[15] Tinetti, M.E. and C.S. Williams, Falls, injuries due to Series. 2006. IOP Publishing.
falls, and the risk of admission to a nursing home. [30] Narayanan, M.R., et al. Falls management: detection
New England journal of medicine, 1997. 337(18): p. and prevention, using a waist-mounted triaxial
1279-1284. accelerometer. in Engineering in Medicine and
[16] Almeida, O., M. Zhang, and J.-C. Liu. Dynamic fall Biology Society, 2007. EMBS 2007. 29th Annual
detection and pace measurement in walking sticks. in International Conference of the IEEE. 2007. IEEE.
High Confidence Medical Devices, Software, and [31] Petelenz, T.J., S.C. Peterson, and S.C. Jacobsen,
Systems and Medical Device Plug-and-Play Elderly fall monitoring method and device. 2002,
Interoperability, 2007. HCMDSS-MDPnP. Joint Google Patents.
Workshop on. 2007. IEEE. [32] Bianchi, F., et al., Barometric pressure and triaxial
[17] Shany, T., et al., Sensors-based wearable systems for accelerometry-based falls event detection. Neural
monitoring of human movement and falls. Sensors Systems and Rehabilitation Engineering, IEEE
Journal, IEEE, 2012. 12(3): p. 658-670. Transactions on, 2010. 18(6): p. 619-627.
[18] Mubashir, M., L. Shao, and L. Seed, A survey on fall [33] Estudillo-Valderrama, M.Á., et al., Design and
detection: Principles and approaches. implementation of a distributed fall detection
Neurocomputing, 2013. 100: p. 144-152. system—personal server. Information Technology in
[19] Arshad, A., et al., A study on health monitoring Biomedicine, IEEE Transactions on, 2009. 13(6): p.
system: recent advancements. IIUM Engineering 874-881.
Journal, 2014. 15(2). [34] Tamura, T., et al., A wearable airbag to prevent fall
[20] Yu, X. Approaches and principles of fall detection injuries. Information Technology in Biomedicine,
for elderly and patient. in e-health Networking, IEEE Transactions on, 2009. 13(6): p. 910-914.

42
 Yoosuf Nizam et al., Int. J. Of Integrated Engineering Vol. 8 No. 1 (2016) p. 35-43

[35] Chen, J., et al. Wearable sensors for reliable fall Journal of Affective Engineering, 2015. 14(1): p. 9-
detection. in Engineering in Medicine and Biology 17.
Society, 2005. IEEE-EMBS 2005. 27th Annual [48] Jansen, B. and R. Deklerck. Context aware inactivity
International Conference of the. 2006. IEEE. recognition for visual fall detection. in Pervasive
[36] Wang, C.-C., et al. Development of a fall detecting Health Conference and Workshops, 2006. 2006.
system for the elderly residents. in Bioinformatics IEEE.
and Biomedical Engineering, 2008. ICBBE 2008. [49] Töreyin, B.U., Y. Dedeoğlu, and A.E. Çetin, HMM
The 2nd International Conference on. 2008. IEEE. based falling person detection using both audio and
[37] Kang, J.M., T. Yoo, and H.C. Kim, A wrist-worn video, in Computer Vision in Human-Computer
integrated health monitoring instrument with a tele- Interaction. 2005, Springer. p. 211-220.
reporting device for telemedicine and telecare. [50] Anderson, D., et al. Recognizing falls from
Instrumentation and Measurement, IEEE silhouettes. in Engineering in Medicine and Biology
Transactions on, 2006. 55(5): p. 1655-1661. Society, 2006. EMBS'06. 28th Annual International
[38] Ghasemzadeh, H., R. Jafari, and B. Prabhakaran, A Conference of the IEEE. 2006. IEEE.
body sensor network with electromyogram and [51] Thome, N. and S. Miguet. A HHMM-based approach
inertial sensors: multimodal interpretation of for robust fall detection. in Control, Automation,
muscular activities. Information Technology in Robotics and Vision, 2006. ICARCV'06. 9th
Biomedicine, IEEE Transactions on, 2010. 14(2): p. International Conference on. 2006. IEEE.
198-206. [52] Miaou, S.-G., F.-C. Shih, and C.-Y. Huang. A smart
[39] Kaluža, B. and M. Luštrek, Fall detection and vision-based human fall detection system for
activity recognition methods for the confidence telehealth applications. in Proc. Third ISATED Int.
project: a survey. Population, 2010. 2020(2030): p. Conf. on Telehealth. 2007.
2040. [53] Miaou, S.-G., P.-H. Sung, and C.-Y. Huang. A
[40] Kangas, M., et al., Sensitivity and specificity of fall customized human fall detection system using omni-
detection in people aged 40 years and over. Gait & camera images and personal information. in
posture, 2009. 29(4): p. 571-574. Distributed Diagnosis and Home Healthcare, 2006.
[41] Nyan, M., F.E. Tay, and E. Murugasu, A wearable D2H2. 1st Transdisciplinary Conference on. 2006.
system for pre-impact fall detection. Journal of IEEE.
biomechanics, 2008. 41(16): p. 3475-3481. [54] Cucchiara, R., A. Prati, and R. Vezzani, A multi‐
[42] Li, Q., et al. Accurate, fast fall detection using camera vision system for fall detection and alarm
gyroscopes and accelerometer-derived posture generation. Expert Systems, 2007. 24(5): p. 334-345.
information. in Wearable and Implantable Body [55] Rougier, C., et al. Fall detection from human shape
Sensor Networks, 2009. BSN 2009. Sixth and motion history using video surveillance. in
International Workshop on. 2009. IEEE. Advanced Information Networking and Applications
[43] Bourke, A.K. and G.M. Lyons, A threshold-based Workshops, 2007, AINAW'07. 21st International
fall-detection algorithm using a bi-axial gyroscope Conference on. 2007. IEEE.
sensor. Medical engineering & physics, 2008. 30(1): [56] Alwan, M., et al. A smart and passive floor-vibration
p. 84-90. based fall detector for elderly. in Information and
[44] Mohd, M. N. H., Kashima M., K. Sato, K., Communication Technologies, 2006. ICTTA'06. 2nd.
Watanabe, M. 2015. A Non-invasive Facial Visual- 2006. IEEE.
Infrared Stereo Vision Based Measurement as an [57] Scott, T.E., Bed exit detection apparatus. 2000,
Alternative for Physiological Measurement. In Google Patents.
Lecture Notes in Computer Science (including [58] Sixsmith, A. and N. Johnson, A smart sensor to
subseries Lecture Notes in Artificial Intelligence and detect the falls of the elderly. Pervasive Computing,
Lecture Notes in Bioinformatics), 9009: 684-697. IEEE, 2004. 3(2): p. 42-47.
[45] Mohd, M. N. H., Kashima M., K. Sato, K., [59] Sixsmith, A., N. Johnson, and R. Whatmore.
Watanabe, M. 2015. Internal state measurement from Pyroelectric IR sensor arrays for fall detection in the
facial stereo thermal and visible sensors through older population. in Journal de Physique IV
SVM classification. ARPN Journal of Engineering (Proceedings). 2005. EDP sciences.
and Applied Sciences, 10(18): 8363-8371. [60] Tabar, A.M., A. Keshavarz, and H. Aghajan. Smart
[46] Mohd, M. N. H., KASHIMA, M., SATO, K., & home care network using sensor fusion and
WATANABE, M. Internal Emotion State (Mental distributed vision-based reasoning. in Proceedings of
Stress) Based on the Information from Thermal and the 4th ACM international workshop on Video
Visible Sensors. surveillance and sensor networks. 2006. ACM.
[47] Mohd, M. N., KASHIMA, M., SATO, K., & [61] Zhuang, X., et al. Acoustic fall detection using
WATANABE, M. Mental Stress Recognition based Gaussian mixture models and GMM supervectors. in
on Non-invasive and Non-contact Measurement from Acoustics, Speech and Signal Processing, 2009.
Stereo Thermal and Visible Sensors. International ICASSP 2009. IEEE International Conference on.
2009. IEEE.

43