scispace - formally typeset
Search or ask a question
Journal ArticleDOI

A fuzzy model for human fall detection in infrared video

TL;DR: The fuzzy model is capable of identifying true and false falls, enhanced with inactivity monitoring aimed at confirming the need for medical assistance and/or care and has been tested for a wide number of static and dynamic falls, demonstrating exciting initial results.
Abstract: Fall detection, especially for elderly people, is a challenging problem which demands new products and technologies. In this paper a fuzzy model for fall detection and inactivity monitoring in infrared video is presented. The classification features proposed include geometric and kinematic parameters associated with more or less sudden changes in the tracked human-related regions of interest. A complete segmentation and tracking algorithm for infrared video as well as a fuzzy fall detection and confirmation algorithm are introduced. The proposed system is capable of identifying true and false falls, enhanced with inactivity monitoring aimed at confirming the need for medical assistance and/or care. The fall indicators used as well as their fuzzy model is explained in detail. The fuzzy model has been tested for a wide number of static and dynamic falls, demonstrating exciting initial results.

Summary (2 min read)

Introduction

  • The thermal infrared imagery is specially suitable for human detection in indoor environments.
  • Then, gradient information from the objects and the background is combined to elaborate a contour map representing the confidence for each pixel to belong to a human’s boundary.

2.1.1. Human candidate blobs detection

  • The algorithm starts with the analysis of an input image I(x, y), captured at time instant t by an infrared camera.
  • The algorithm uses a threshold to perform a binarization for the aim of isolating the human candidates spots.
  • Thus, the warmer zones of the image are isolated, assuming that they belong to humans.
  • These operations require structuring elements that in both cases are 3 × 3 square matrices.
  • Therefore, the algorithm processes each detected blob separately.

2.1.2. Human candidate blobs refinement

  • The first threshold indicates the presence of the head of a human candidate on the current column, which the authors use as indicator of a local maximum as heads are normally warmer than the rest of the human’s body covered by clothes.
  • Once the subROIs have been obtained at the end of the previous section, the authors want to fit the height of each one of them to the real height of the human contained.
  • These sub-ROIs are scanned separately in the next stage.
  • His results are considered a standard in this area.

2.2.1. Fall detection algorithm

  • Firstly, let us explain in more detail the proposed human fall detection algorithm.
  • In their proposal, the general idea of fall detection is based on the joint static and kinematic analysis of the blobs corresponding to humans.
  • In the same sense, let xtr,1 and xdl,1 be the top-right and down-left X-coordinates of the blob in the first image, and let xtr,n and xdl,n be the top-right and down-left corner X-coordinates of the blob in the n-th image.
  • The linguistic variable “HeightChange” is calculated as the relation of the initial height to the final height of the blob.

2.2.3. Inactivity monitoring algorithm

  • Notice that there are two main variables for final decision making.
  • The possible values for Assistance are “AssistanceRequired” and “NoAssistanceRequired”, being initially provided value “NoAssistanceRequired”.
  • If a person has fallen (that is, Alarm has value “FallDetected”) and inactivity monitoring has been active for time interval tm, Assistance is provided the value “AssistanceRequired”.
  • The experiments described next were set up in order to validate the proposed approach.
  • The video sequences were recorded with a FLIR A320 infrared camera at a resolution of 720 × 480 pixels.

3.1. Detection of static falls

  • Characterized through high horizontal velocity and height change.
  • Next, “VelocityChange” reduces to SMALL (0.28), and “HeightChange” decreases to MEDIUM (2.10) terms, which makes change values of variables Alarm and Assistance to “NoFallDetected” and “NoAssistanceRequired”.
  • The second static fall is a backward fall from a “sitting position” (see Fig. 5, third row).
  • This way, Alarm is set to “FallDetected”.

3.3. Identification of false falls

  • False falls are represented in this case study with the kneeling action.
  • In fact, a false fall may be detected as a fall which lasts more than a fall time as well as having small values for fall detection parameters given in the last case shown in Table 2.
  • The proposal described in [24] achieves results with sensitivity and specificity of 90.7% and 90.6%, respectively.
  • The limited number of fall parameters to evaluate further calculations reduces the need for great computation resources.
  • The proposed algorithms are incorporated into a fall detection system for the elderly, and then tested for a wide number of static and dynamic falls, including the identification of false falls.

Did you find this useful? Give us your feedback

Content maybe subject to copyright    Report

AUTHOR COPY
Journal of Intelligent & Fuzzy Systems 24 (2013) 215–228
DOI:10.3233/IFS-2012-0548
IOS Press
215
A fuzzy model for human fall detection
in infrared video
Marina V. Sokolova
a
, Juan Serrano-Cuerda
b
, Jos
´
e Carlos Castillo
b
and Antonio Fern
´
andez-Caballero
b,
a
South-West State University, Kursk, Russia
b
Instituto de Investigaci´on en Inform´atica de Albacete, Universidad de Castilla-La Mancha, Albacete, Spain
Abstract. Fall detection, especially for elderly people, is a challenging problem which demands new products and technologies.
In this paper a fuzzy model for fall detection and inactivity monitoring in infrared video is presented. The classification features
proposed include geometric and kinematic parameters associated with more or less sudden changes in the tracked human-related
regions of interest. A complete segmentation and tracking algorithm for infrared video as well as a fuzzy fall detection and
confirmation algorithm are introduced. The proposed system is capable of identifying true and false falls, enhanced with inactivity
monitoring aimed at confirming the need for medical assistance and/or care. The fall indicators used as well as their fuzzy model
is explained in detail. The fuzzy model has been tested for a wide number of static and dynamic falls, demonstrating exciting
initial results.
Keywords: Fall detection, video segmentation and tracking, infrared video, fuzzy system
1. Introduction
Nowadays, fall detection is still a challenging and
emergent problem [1], especially for monitoring elderly
people [8, 13]. The problem has mainly been provoked
and induced by population ageing, showing a tendency
of permanent growth in accordance with recent demo-
graphical predictions [29]. As reported, about a third
part of humans aged 65 and above suffer from a fall
each year, increasing up to 42 percent for elderly over
70 years. Moreover, every year approximately around
50 percent of humans who live in long-term care insti-
tutions suffer a fall, and 40 percent of them experience
recurrent falls. In addition to this, elderly people are
highly vulnerable, as falls are known to be one of the
leading causes of injuries and death. Another important
consequence of recurrent falls is the post-fall syndrome
Corresponding author. Antonio Fern
´
andez-Caballero, Instituto
de Investigaci
´
on en Inform
´
atica de Albacete, Universidad de Castilla-
La Mancha, 02071 - Albacete, Spain. Email: Antonio.Fdez@uclm.es.
that manifests through depression, loss of autonomy,
immobilization, and may result in impairments in daily
activities [29].
One more reason of growing research in the field of
fall detection is that governmental promotion through
high public attention, and constant improvement of
associated support technology in the area of elderly
care, permits to carry out research and to implement
new applications. This enables usage of professional
up-to-date equipment and development of complex
solutions based on advanced visual information analy-
sis methods (e.g. [3, 17]). Consequently, the number of
applications in this area has constantly grown. Indeed,
in [31] an overview of the current situation in the
field of elderly fall detection and alarm generation
is presented. Firstly, the authors classify fall detec-
tion under camera-based, wearing-device-based and
ambient-device-based approaches. In another review
on the principles and algorithms for fall detection
[21], the authors study analytical methods and machine
learning techniques. Within the first group of meth-
ods, image processing and usage of sensors (horizontal
1064-1246/13/$27.50 © 2013 IOS Press and the authors. All rights reserved

AUTHOR COPY
216 M.V. Sokolova et al. / A fuzzy model for human fall detection in infrared video
inclination sensor, sensitive floor tiles, accelerome-
ters, etc.) are included. In a second group, the authors
place approaches based on machine learning for fall
detection, covering all steps from observation to clas-
sification.
Within the camera-based group of methods, several
approaches to fall detection have been proposed. For
instance, head motion trajectory analysis is one of the
indicators of a fall [23]. Many research works are ded-
icated to posture detection analysis with the intention
not only to identify poses, but also to distinguish among
true and false falls. In this way, in [28] the authors pro-
pose a 3D method from 2D pose reconstruction based
on the introduction of bone, bone symmetry, and rigid
body projection constraints. Another approach for fall
detection, which is based on the k-nearest neighbor
algorithm, is introduced in [15]. Moreover, 3D shape-
encoded filters are described in [19]. These extract
the boundary gradient information of the body image,
which is then propagated by a branching particle sys-
tem. In [25], the application of a kernel particle filter
for 3D body tracking is introduced. In [18] a sys-
tem for human behavior analysis is reported, where a
human is represented with three components: human
star skeleton, angles of six sticks in the star skeleton, and
object motion vectors. Here the human posture classi-
fier, based on a multi-category support vector machine,
is used for pose detection.
In general terms, human fall detection starts with peo-
ple monitoring and tracking and finishes with secure
decision making on whether a human has fallen or
not. It is a known fact that segmentation and tracking
results are very sensitive to environmental conditions.
The results are dependent on the quality of the image, on
sudden changes in illumination (mainly in color video)
and temperature (in infrared video), among others. In
addition, humans have different types of body composi-
tion, which can be heavy, sturdy, slim, stocky, and so on.
In 2-D images, humans can be overshadowed by other
objects and be part of groups composed of two, three or
more humans. Thereby, direct analysis of a human body
appears to be an ambiguous and difficult task. Indeed,
on the one hand, the principal anatomical proportions
of human bodies are similar. But, also, personal phys-
ical characteristics of each human make it impossible
to create a uniform crisp decision block. Thus, since
humans usually appear warmer than the background in
many environmental conditions, several methods which
use thermal infrared imagery have been proposed for
human detection [5, 11]. The thermal infrared imagery
is specially suitable for human detection in indoor
environments. These are the most usual areas of
work for human fall detection. Moreover, temperature
changes affecting the detection do not usually appear
in indoor, since there are few external factors which
can affect them (e.g., air currents and/or day/night heat
variations).
There have been several approaches to infrared
human detection so far [12]. For instance, a pedes-
trian detection method in thermal infrared imagery
using the wavelet transform is proposed in [14]. The
approach uses a support vector machine (SVM) clas-
sifier to establish true pedestrian regions. The regions
are detected using features obtained from a conversion
to the frequency domain used as an input to the SVM.
In [7], a background subtraction is initially performed
to identify local foreground objects. Then, gradient
information from the objects and the background is
combined to elaborate a contour map representing the
confidence for each pixel to belong to a human’s bound-
ary. Finally, a path search is performed to complete
broken segments. An approach based on histogram is
introduced on [10], realizing a vertical and a horizon-
tal adjustment on the pedestrian candidate regions. Two
different cases for summer and winter are established,
using the gray level values to delimit pedestrians in
the winter case and the intensity changes in the human
boundaries to delimit pedestrians in summer sequences.
More recently, the Otsu algorithm [22] is used to bina-
rize the image separating the warmer regions (probably
belonging to humans) from the colder ones, using open-
ing and closing morphological operations to remove
small broken parts obtained from the binarization [6].
Next, objects that are too small or have a wrong aspect
ratio to be humans are filtered out and a histogram of
oriented gradients (HOG) method is used to extract
features used to train an Adaboost classifier.
In this paper, we explain how human fall detection
and inactivity monitoring in infrared video is real-
ized within a fuzzy model. Fuzziness has been largely
studied in the past for pattern recognition [9, 27]. In
relation to fall detection, a method with fuzzy one class
support vector machine in a smart room has been pre-
sented [30]. The fuzzy membership function represents
the likelihood for the data to belong to a target class
[20]. This is also our approach. The paper is orga-
nized as follows. Section 2 introduces the proposed
approach towards human monitoring and fall detec-
tion, mainly explaining the required segmentation and
fall detection algorithms. Then, in Section 3, data and
experimental results are presented. Section 4 introduces
some aspects of the accuracy of the proposed fuzzy

AUTHOR COPY
M.V. Sokolova et al. / A fuzzy model for human fall detection in infrared video 217
system and shows some comparison with other similar
approaches. Finally, some final conclusions are pro-
vided in Section 5.
2. Human monitoring for fall detection
Our approach is based on the idea of combining a fall
detection with an inactivity monitoring algorithm. In
this case, the inactivity monitoring subsystem is thought
for checking if a human continues laying inactive during
some predefined time after a fall has been detected in the
detection subsystem. In this paper, fall detection as such
is based on the geometrical analysis of the segmented
region of interest (ROI) representing a human. Roughly,
it is decided that a fall has occurred through the study
of the velocity of the deformation suffered by the ROI
along an established time interval. This time interval is
denominated the “Fall time”, t
f
, which typically takes
values between 1 and 3 seconds. Additionally, fuzzy
logic is used in order to make decisions more flexi-
ble, to avoid some limitations in the representation of
human figures, and to smooth the limits of the evalu-
ation parameters. The objectives of our fall detection
and inactivity monitoring approach are the following
ones:
to detect static falls (of stationary humans) from a
standing or sitting position,
to detect dynamic falls (of walking or running
humans), which may be frontal, backward and lat-
eral (to the right or to the left) falls, and,
to identify false falls such as kneeling and bending.
Lastly, in case of severe falls, when the human does
not stand up after some pre-established inactivity mon-
itoring time, t
m
, the urgent need for medical assistance
and/or care is assessed.
2.1. Segmentation for human detection
The proposed human detection algorithm is
grounded on a segmentation algorithm based on a single
frame. The algorithm is explained in detail in the fol-
lowing sections related to the different phases, namely,
human candidate blobs detection, human candidate
blobs refinement and human confirmation.
2.1.1. Human candidate blobs detection
The algorithm starts with the analysis of an input
image I(x, y), captured at time instant t by an infrared
camera. Firstly, a change in scale, as shown in Equation
(1) is performed in order to normalize the image. The
objective of this operation is to always work on a similar
scale of values, transforming I(x, y)toI
1
(x, y). The
normalization assumes a factor γ experimentally fixed
as 60 and uses the mean gray level value of the current
image,
¯
I.
I
1
(x, y) =
I(x, y) × γ
¯
I
(1)
where I
1
(x, y) is the normalized image. Notice that
I
1
(x, y) = I(x, y) when
¯
I = γ.
The algorithm uses a threshold to perform a bina-
rization for the aim of isolating the human candidates
spots. The threshold is calculated using adaptive thresh-
olding [26] based on the standard deviation of I(x, y),
that is, σ
I(x,y)
, obtaining the image areas which contain
moderate heat blobs, and, therefore, belong to human
candidates. Thus, the warmer zones of the image are
isolated, assuming that they belong to humans. As the
image has been scaled, the threshold θ
c
is calculated as:
θ
c
=
5
4
(γ + σ
I
1
)(2)
where σ
I
1
is the standard deviation of image I
1
(x, y).
Notice that a tolerance value of a 25% above the sum
of the mean image gray level value and the image gray
level value standard deviation is offered.
Now, image I
1
is binarized using the obtained thresh-
old θ
c
. Pixels above the threshold are set as maximum
gray level value max = 255 and pixels below are set
as minimum gray level value min = 0 as shown in
Equation (3).
I
1
b
(x, y) =
min, if I
1
(x, y) θ
c
max, otherwise
(3)
Next, morphological opening and closing operations
are performed to eliminate isolated pixels and to unite
areas split during the binarization. These operations
require structuring elements that in both cases are 3 ×3
square matrices.
Afterwards, the blobs contained in the image are
obtained. A minimum area (around
1
16
of the image size)
is established for consider a blob to contain a human.
As a result, a list of blobs, L
B
, containing human
candidates in forms b
λ
[(x
start
,y
start
), (x
end
,y
end
)] is
generated. λ stands for the number of the human can-
didate blobs in image I
1
(x, y), whereas (x
start
,y
start
)
and (x
end
,y
end
) are the upper left and lower right
coordinates, respectively, of the minimum rectangle
containing the blobs. At this point, there is a need to

AUTHOR COPY
218 M.V. Sokolova et al. / A fuzzy model for human fall detection in infrared video
validate the content of each blob to find out if it con-
tains one single human candidate or more than one.
Therefore, the algorithm processes each detected blob
separately.
Let us define a region of interest (ROI) as the min-
imum rectangle containing one blob of a list L
B
.
A ROI may be defined as R
λ
= R
λ
(i, j) when asso-
ciated to blob b
λ
[(x
start
,y
start
), (x
end
,y
end
)]. Notice
that i[1..max
i
= x
end
x
start
+ 1] and j[1..max
j
=
y
end
y
start
+ 1].
2.1.2. Human candidate blobs refinement
2.1.2.1. Human vertical delimiting. The first step con-
sists in scanning R
λ
by columns, adding the gray level
value corresponding to each column pixel, as shown in
equation (4):
H
λ
[i] =
max
j
j=1
R
λ
(i, j), i[1..max
i
](4)
This way, a histogram H
λ
[i], showing which zones
of the ROI own greater heat concentrations, is obtained
in order to increase the certainty of the presence and
situation of human heads, as well as separate human
groups (if any) into single human. For this purpose,
local maxima and local minima are searched in H
λ
[i]
to establish the different heat sources. To assess whether
a histogram column contains a local maximum or min-
imum, a couple of thresholds are fixed, θ
v
max
, and θ
v
min
.
The first threshold indicates the presence of the head
of a human candidate on the current column, which
we use as indicator of a local maximum as heads are
normally warmer than the rest of the human’s body cov-
ered by clothes. On the other hand, θ
v
min
indicates those
regions of the ROI where the sum of the heat sources
are really low. These regions are supposed to belong to
gaps between two humans and whose column summed
gray level is below a 30% of the mean ROI gray level
value. These two thresholds are calculated as shown in
Equations (5) and (6) respectively.
θ
v
max
= (
4
3
λ) + σ
I
1
(5)
θ
vmin
= 0.3 × R
λ
× max
j
(6)
At the end of this process we obtain the human can-
didates of the scene separated into sub-ROIs, sR
λ,α
,
vertically adjusted.
2.1.2.2. Human horizontal delimiting. Once the sub-
ROIs have been obtained at the end of the previous
section, we want to fit the height of each one of them to
the real height of the human contained. This adjustment
is performed by applying a new threshold, θ
h
. The cal-
culation is applied separately on each sub-ROI to avoid
the influence of the rest of the image on the result. This
threshold takes the value of the sub-ROI average gray
level, θ
h
= sR
λ,α
. Thus, sub-ROI sR
λ,α
is binarized to
delimit its upper and lower limits, obtaining sR
λ,α
as
shown in Equation (7) similarly to Equation (3.)
sR
λ,α
(i, j) =
min, if sR
λ,α
(i, j) θ
h
max, otherwise
(7)
After this, a closing is performed to unite spots iso-
lated in the binarization, getting sR
λ,α
. Next, sR
c,λ
is
scanned, searching pixels with values superior to min.
The upper and lower rows of the human candidate are
equal to the first and last rows, respectively, containing
pixels with a value set to max. This way, we obtain a new
set of sub-ROIs sR
λ,α
. These sub-ROIs are scanned
separately in the next stage.
2.1.2.3. Human confirmation. Now a final stage is
needed for each sub-ROI, sR
λ,α
, to confirm if the
human candidate contained in it is actually a human.
At this point, it is interesting to remind that every
sub-ROI is defined by its coordinates (x
start
,y
start
) and
(x
end
,y
end
). In first place, let us define the basic param-
eters needed for human confirmation for every sub-ROI.
These are the “height of the sub-ROI”, h
sR
, the “width
of the sub-ROI”, w
sR
, and the “height-to-width of the
sub-ROI”, R
hw
sR
. The parameters are calculated as
described next:
h
sR
= x
end
x
start
(8)
w
sR
= y
end
y
start
(9)
R
hw
sR
=
h
sR
w
sR
(10)
In order to confirm the presence of a human in a
sub-ROI, the criterium established by Loomis [16] is
used as an inspiration. Loomis studied in the 1940’s the
body part proportions of humans for different builds.
His results are considered a standard in this area. Exper-
imentally, and following his recommendations, we have
imposed that R
hw
sR
[0.2 ...7.0]. Finally, the blobs
associated to ROIs that have not satisfied this criterium
are deleted from the original blobs list, L
B
.

AUTHOR COPY
M.V. Sokolova et al. / A fuzzy model for human fall detection in infrared video 219
2.2. Fuzzy-based fall detection algorithm
Our approach is based on the idea of combining a fall
detection and an inactivity monitoring algorithm. In this
case, the inactivity monitoring algorithm is thought for
checking if a human continues laying inactive during
some predefined time after a fall has been detected in the
detection subsystem. In this paper, fall detection as such
is based on the geometrical analysis of the blobs listed
in L
B
and representing humans. Roughly, it is decided
that a fall has occurred through the study of the veloc-
ity of the deformation suffered by the blob along an
established time interval. This time interval is denomi-
nated the “Fall time”, t
f
, which typically takes values
between 1 and 3 seconds. Additionally, fuzzy logic is
used in order to make decisions more flexible, to avoid
some limitations in the representation of human figures,
and to smooth the limits of the evaluation parameters.
Figure 1 illustrates the principal ideas of our
approach towards fall detection and inactivity monitor-
ing. In brief, during “Fall time”, t
f
, the fall detection
subsystem is launched (“A in Fig. 1). Once a fall
is detected, the inactivity monitoring subsystem is
activated (“B” and “C” in Fig. 1). If the fallen human
stands up before a pre-established time interval,
denominated “Monitoring time”, t
m
(typically 30
seconds), the inactivity monitoring subsystem is
stopped, and the fall detection subsystem is launched
again (“D” and “E” in Fig. 1).
2.2.1. Fall detection algorithm
Firstly, let us explain in more detail the pro-
posed human fall detection algorithm. In our proposal,
the general idea of fall detection is based on the
joint static and kinematic analysis of the blobs
corresponding to humans. The statical analysis cor-
responds to the study of a single (and current input)
segmented infrared image, whilst kinematic analy-
sis is based on a sequence of segmented infrared
video images. During statical analysis, the geometrical
parameters of each present blob are calculated, namely,
the “change of the height of the blob”, h, and the
“width-to-height ratio of the blob”, R
wh
. The kinematic
analysis implies the calculation of the “velocity of the
change of the height of the blob”, v, among a series
of n consecutive segmented video frames. All these
parameters, together with the predefined “Fall time”,
t
f
, compose a set of fall indicators used in this research.
2.2.1.1. Calculation of the fall indicators. For the n
consecutive blobs corresponding to a same human,
let y
tr,1
and y
dl,1
be the top-right and down-left Y-
coordinates of the blob in the first image, and let
y
tr,n
and y
dl,n
be the top-right and down-left corner
Y-coordinates of the blob in the last (n-th) image,
respectively. In the same sense, let x
tr,1
and x
dl,1
be
the top-right and down-left X-coordinates of the blob
in the first image, and let x
tr,n
and x
dl,n
be the top-right
and down-left corner X-coordinates of the blob in the
n-th image. Let us also define x
tr
, x
dl
, y
tr
and y
ld
as
the top-right and down-left coordinates of the current
blob. Also, let w = x
tr
x
dl
and h = y
tr
y
dl
be the
width and the height, respectively, of the current blob.
The parameters are calculated as follows:
R
wh
=
x
tr
x
dl
y
tr
y
dl
(11)
h =
y
tr,1
y
dl,1
y
tr,n
y
dl,n
(12)
Fig. 1. Graphical representation of the fall detection and inactivity monitoring system.

Citations
More filters
Journal ArticleDOI
TL;DR: It is shown that a k-nn classifier is competitive on the publicly available URFD dataset in terms of sensitivity and specificity while being much more simple to implement on an embedded platform.

126 citations


Cites background from "A fuzzy model for human fall detect..."

  • ...Thermal imaging cameras, also called infrared cameras, which detect the heat given off by an object or human can also deliver very valuable source of information for detecting falls [11]....

    [...]

Journal ArticleDOI
01 Mar 2016
TL;DR: A new application of fuzzy logic in a novel approach to modeling and reliable low cost detecting of falls is presented.
Abstract: Graphical abstractDisplay Omitted HighlightsA new approach for reliable fall detection.In case of potential fall a threshold-based algorithm launches the fuzzy system to authenticate the fall event. The fuzzy system consists of two input Mamdani engines and a triggering alert Sugeno engine.The output of the first engine is a fuzzy set, which assigns grades of membership to the possible values of dynamic transitions, whereas the output of the second one is another fuzzy set assigning membership grades to possible body poses.Since the Mamdani engines perform fuzzy reasoning on disjoint subsets of the linguistic variables, the total number of the fuzzy rules needed for input-output mapping is far smaller. In this paper, we present a new approach for reliable fall detection. The fuzzy system consists of two input Mamdani engines and a triggering alert Sugeno engine. The output of the first Mamdani engine is a fuzzy set, which assigns grades of membership to the possible values of dynamic transitions, whereas the output of the second one is another fuzzy set assigning membership grades to possible body poses. Since Mamdani engines perform fuzzy reasoning on disjoint subsets of the linguistic variables, the total number of the fuzzy rules needed for input-output mapping is far smaller. The person pose is determined on the basis of depth maps, whereas the pose transitions are inferred using both depth maps and the accelerations acquired by a body worn inertial sensor. In case of potential fall a threshold-based algorithm launches the fuzzy system to authenticate the fall event. Using the accelerometric data we determine the moment of the impact, which in turn helps us to calculate the pose transitions. To the best of our knowledge, this is a new application of fuzzy logic in a novel approach to modeling and reliable low cost detecting of falls.

56 citations


Cites methods from "A fuzzy model for human fall detect..."

  • ...For that reason, in [15][2] in order to recognize different activities in various environments, both controlled as well as unstructured, an infrared illumination was utilized to enable the web cameras to deliver images of sufficient quality in poor lighting conditions....

    [...]

  • ...Several methods have been developed so far to detect falls using various kinds of video cameras [15][2][16]....

    [...]

Journal ArticleDOI
TL;DR: Experimental results along with the collected datasets and open database showed that the proposed method achieved higher accuracy of behavior recognition when compared to conventional methods.
Abstract: Our method can show plural recognition results, i.e. person is waving while walking.We propose behavior recognition fusing current recognition and prediction of behavior.For the fusion, the fuzzy system based classifier of behaviors is adopted. With the development of intelligent surveillance systems, human behavior recognition has been extensively researched. Most of the previous methods recognized human behavior based on spatial and temporal features from (current) input image sequences, without the behavior prediction from previously recognized behaviors. Considering an example of behavior prediction, punching is more probable in the current frame when the previous behavior is standing as compared to the previous behavior being lying down. Nevertheless, there has been little study regarding the combination of currently recognized behavior information with behavior prediction. Therefore, we propose a fuzzy system based behavior recognition technique by combining both behavior prediction and recognition. To perform behavior recognition during daytime and nighttime, a dual camera system of visible light and thermal (far infrared light) cameras is used to capture 12 datasets including 11 different human behaviors in various surveillance environments. Experimental results along with the collected datasets and open database showed that the proposed method achieved higher accuracy of behavior recognition when compared to conventional methods.

46 citations

Journal ArticleDOI
10 Apr 2014-Sensors
TL;DR: This paper investigates the robustness of a new thermal-infrared pedestrian detection system under different outdoor environmental conditions and draws firm conclusions about the conditions under which it can be affirmed that it is efficient to use the proposal to robustly extract human ROIs.
Abstract: This paper investigates the robustness of a new thermal-infrared pedestrian detection system under different outdoor environmental conditions. In first place the algorithm for pedestrian ROI extraction in thermal-infrared video based on both thermal and motion information is introduced. Then, the evaluation of the proposal is detailed after describing the complete thermal and motion information fusion. In this sense, the environment chosen for evaluation is described, and the twelve test sequences are specified. For each of the sequences captured from a forward-looking infrared FLIR A-320 camera, the paper explains the weather and light conditions under which it was captured. The results allow us to draw firm conclusions about the conditions under which it can be affirmed that it is efficient to use our thermal-infrared proposal to robustly extract human ROIs.

34 citations


Cites methods from "A fuzzy model for human fall detect..."

  • ...This paper introduces a new algorithm for robust ROI extraction of pedestrians in thermal-infrared video based on the authors’ previous works [16,17]....

    [...]

Journal ArticleDOI
Min Li1, Guanghua Xu1, Bo He1, Xiaolong Ma1, Jun Xie1 
TL;DR: In this article, the authors defined a dynamic supporting area containing both feet and the area between the two feet, and proposed a method of fall prediction based on a modified zero moment point criterion using motion-monitoring data from a Kinect sensor.
Abstract: Accidental falls have always been a serious problem for the elderly. There is considerable demand for pre-impact fall detection systems with long lead times. According to the zero moment point criterion, the zero moment point should be kept beneath the supporting foot for stability during humanoid robot standing or walking. However, the zero moment point in the human walk does not stay fixed under the supporting foot. In this paper, we define a dynamic supporting area containing both feet and the area between the two feet, and propose a method of fall prediction based on a modified zero moment point criterion using motion-monitoring data from a Kinect sensor. A fall event is predicted if the projection of the zero moment point locates outside of the dynamic supporting area. The proposed method is compared with a method identifying the imbalance state based on a support vector machine classifier. Experimental results show that fall events could be detected with an average lead time of 867.9 ms (SD = 199.2), a sensitivity of 100%, a specificity of 81.3%, a positive predictive value of 87.0%, a negative predictive value of 100%, and an accuracy of 91.7% using the modified zero moment point criterion. The lead time was 571.9 ms (SD = 153.5) and accuracy was 100% for the support vector machine classifier. The modified zero moment point criterion-based method achieved the longest lead time in the pre-impact fall detection.

32 citations


Cites background or methods from "A fuzzy model for human fall detect..."

  • ...or tilt sensors) [2], cameras [3], vibration sensors to detect floor vibration or sound caused by falls [4], and smart-...

    [...]

  • ...Inactivity/change of shape [8], head motion trajectory analysis [9], and posture detection analysis [3] are often used for camera-based fall detection approaches....

    [...]

References
More filters
Journal ArticleDOI

37,017 citations


"A fuzzy model for human fall detect..." refers methods in this paper

  • ...More recently, the Otsu algorithm [22] is used to binarize the image separating the warmer regions (probably belonging to humans) from the colder ones, using opening and closing morphological operations to remove small broken parts obtained from the binarization [6]....

    [...]

Journal Article
TL;DR: How the field of computer (and robot) vision has evolved, particularly over the past 20 years, is described, and its central methodological paradigms are introduced.

3,112 citations


"A fuzzy model for human fall detect..." refers methods in this paper

  • ...The threshold is calculated using adaptive thresholding [26] based on the standard deviation of I(x, y), that is, σI(x,y), obtaining the image areas which contain moderate heat blobs, and, therefore, belong to human candidates....

    [...]

Book
02 Feb 2001
TL;DR: Computer Vision presents the necessary theory and techniques for students and practitioners who will work in fields where significant information must be extracted automatically from images, a useful resource book for professionals and a core text for both undergraduate and beginning graduate computer vision and imaging courses.
Abstract: From the Publisher: Computer Vision presents the necessary theory and techniques for students and practitioners who will work in fields where significant information must be extracted automatically from images. It will be a useful resource automatically from images. It will be a useful resource book for professionals and a core text for both undergraduate and beginning graduate computer vision and imaging courses. Features Topics include image databases an virtual and augmented reality in addition to classical topics. Offers a complete view of two real-world systems that use computer vision. Contains applications from industry, medicine, land use, multimedia, and computer graphics. Includes over 250 exercises and programming projects, 48 separately defined algorithms, and 360 figures. The companion website features include image archive, sample

1,880 citations


"A fuzzy model for human fall detect..." refers methods in this paper

  • ...The threshold is calculated using adaptive thresholding [26] based on the standard deviation of I(x, y), that is, σI(x,y), obtaining the image areas which contain moderate heat blobs, and, therefore, belong to human candidates....

    [...]

Proceedings ArticleDOI
22 Oct 2007
TL;DR: The difficulty to compare the performances of the different systems due to the lack of a common framework is pointed out and a procedure for this evaluation is proposed.
Abstract: Fall detection of the elderly is a major public health problem. Thus it has generated a wide range of applied research and prompted the development of telemonitoring systems to enable the early diagnosis of fall conditions. This article is a survey of systems, algorithms and sensors, for the automatic early detection of the fall of elderly persons. It points out the difficulty to compare the performances of the different systems due to the lack of a common framework. It then proposes a procedure for this evaluation.

581 citations


"A fuzzy model for human fall detect..." refers methods in this paper

  • ...In another review on the principles and algorithms for fall detection [21], the authors study analytical methods and machine learning techniques....

    [...]

01 Jan 2007

263 citations


"A fuzzy model for human fall detect..." refers result in this paper

  • ...The problem has mainly been provoked and induced by population ageing, showing a tendency of permanent growth in accordance with recent demographical predictions [29]....

    [...]

Frequently Asked Questions (1)
Q1. What have the authors contributed in "A fuzzy model for human fall detection in infrared video" ?

In this paper a fuzzy model for fall detection and inactivity monitoring in infrared video is presented. The proposed system is capable of identifying true and false falls, enhanced with inactivity monitoring aimed at confirming the need for medical assistance and/or care.