Multicamera video surveillance fall detection system for elders
K. ARUL SELVAN
Asst. Professor, Department of Instrumentation and Control Engineering, A.V.C College of
Engineering
J.R. MAHA LAKSHMI
III - Year, Department of Instrumentation and Control Engineering, A.V.C College of Engineering
maha77249@[Link]
P. JAMUNA SRI
III - Year, Department of Instrumentation and Control Engineering, A.V.C College of Engineering
pvjr2005@[Link]
S. SHIVA SREE
III - Year, Department of Instrumentation and Control Engineering, A.V.C College of Engineering
shivasreesenguttuvan@[Link]
Abstract elderly population and the most likely cause of
Video Surveillance is an omnipresent accidental death.
topic when it comes to enhancing security and
safety in the intelligent home environment. According to the Public Health
Artificial vision provides a remarkable good Agency, one person out of eight was older
sensor. Cameras are passive sensors that than 65 years old in 2001. In 2026, this
supply a great amount of [Link] this proportion will be one out of five. Moreover, a
project we develop an application for elderly majority of seniors, 93%, reside in private
care that detects falls or faints and house, and among them, 29% live alone. Falls
automatically triggers the health alarm. In this are one of the major risks for old people living
work, we propose a human-shape-based falling alone, causing severe [Link] gravity of
detection algorithm and implement this the situation can increase if the person cannot
algorithm in a multi-camera video surveillance call for help, being unconscious or
system. This algorithm uses multiple cameras immobilized. Nowadays, the usual solution to
to fetch the images from different regions detect falls is to use some wearable sensors
required to monitor. It then uses a falling- like accelerometers or help buttons. However,
pattern recognition approach to determine if an older people often forget to wear them, and in
accidental falling has ccurred. If yes, the the case of a help button, it is useless if the
system will trigger the health alarm. It should person is unconscious after the fall.
not reset within few seconds system
automatically sends short messages to Moreover, batteries are needed for
someone needs to alert. Furthermore, we use these devices and must be replaced or
the speed of fall to differentiate real fall recharged regularly for adequate functioning.
incident and an event where the person is
simply lying without falling.
[Link] of the Project
Keywords- passive sensors, video
surveillance,pattern recognition, speed of fall The objective of this project is to
detect the accidental falls of elderly people
I. Introduction using video monitoring. A computer with dual
Frequency of falls amongst elderly web camera is going to monitor the elderly
people is high. About 30% of independent people.
living elderly persons (>65yr) make at least
one fall a year. Half of those over 80 fall at The frames are analyzed using the
least once a year. About 10% to 15% of all prescribed algorithm, if any fall is detected the
falls in older people will result in some serious system will raise alarm. The alarm will
physical injury, e.g. fractures (in 5% of all produce sounds for few minutes to indicate fall
falls). Fall-related injuries are reported to be is happened. If it is not reset manually within a
the fifth most common cause of death in the preset time, it resets automatically and fall
information is transferred to the care taker via The first step in our method is to
short message through mobile phone. obtain the segmentation of the foreground
human object. We achieve this by adapting
background subtraction algorithm developed
by Mohamed sigari [5].
The next step is to skeletonise the
[Link] Works human object in order to maintain their
privacy. From the silhouette of the foreground
Recently some research has been done human object thinning algorithm is applied.
to detect falls using image processing Using the skeleton fall is detected by
techniques. A simple method was used in [1] projection histogram Àpar method [6]. With
based on analyzing aspect ratio of the moving the help of the height and width of the
object’s bounding box. This method could be skeleton, projection histogram is plotted.
inaccurate, depending on the relative position Using the plot of the projection histogram fall
of the person, camera, and perhaps occluding is identified.
objects. The works in [2] used the normalized
vertical and horizontal projection of Finally we use the adjacent frame
segmented object as feature vectors. To difference method to identify the falling speed
overcome occluding objects problem, some and to infer the real falling events. These
researchers have mounted the camera on the algorithm steps are portrayed in figure.1.
ceiling: Lee [3] detected a fall by analyzing the
shape and 2D velocity of the person. Nait-
Charif [4] tracked the person using an ellipse
and inferring falling incident when target
person is detected as inactive outside normal
zones of inactivity like chairs or sofas.
Despite the considerable achievements
that has accomplished on this field in the
recent years, there are still some clear
challenges to overcome:
• Visual fall detection is inherently prone to
high levels of false positive as what appears to
be a fall or not be a fall. In other words, most
of current systems are unable to discriminate
between real fall incident and an event when
person is lying or sitting down abruptly.
• Existent fall detection systems tend to deal
with restricted movement patterns and fall
incidents are usually detected in contrast with
limited normal scenarios like walking; Figure [Link] flow diagram
however in real home environments various
normal /abnormal motions occur. We will now discuss the various algorithms
used in the proposed system.
[Link] System Overview
A) Foreground Segmentation
In this paper, we consider indoor
environment setting with dual camera. We Identifying moving objects from a
assume camera point of view is easily video sequence is a fundamental and critical
recognising human postures without task in many computer-vision applications. A
ambiguities. common approach is to perform background
subtraction, which identifies moving objects
from the portion of a video frame that differs
significantly from a background model. There
are many challenges in developing a good
background
subtraction algorithm.
First, it must be robust against changes
in illumination. Second, it should avoid
detecting non-stationary background objects
such as moving leaves, rain, snow, and
shadows cast by moving objects. Finally, its
internal background model should react
quickly to changes in background such as
starting and stopping of vehicles.
Background subtraction is particularly
popular method for motion segmentation. The
traditional background extraction methods
such as average method, histogram method,
GMM commonly could not precisely extract Figure 2 (b) Background image
background when moving objects have high with foreground object
density or appear too frequently. In order to
solve this problem, popular method for motion
segmentation is used and attempts to detect
moving regions in an image by differencing
between current image and a reference
background image in a pixel-by pixel
fashion .This method extracts the foreground
object shown in figure 2a, b.c which is fairly
robust and gives appropriate results on image
sequences with shadows, highlights.
Figure 2 (c) Background
subtracted image
B) Silhouette Formation and
Skeletonaisation
Complex models and scenes are
rendered as simple line drawings by rendering
silhouette edges. This definition includes
internal silhouettes as well as the object’s
outline, or halo. It is important to note that the
Figure 2 (a) Background image
silhouette set of an object is view dependent,
that is the edges of a model that are silhouettes
change based on the point from which the
object is viewed. Here silhouette of the For this contest, we directly used the
foreground image is obtained with the help of foreground images. The feature vectors are
the method developed by Bruce grooch [7] and then obtained from the projection histograms
is shown in figure 3. of the foreground mask i.e. projections of the
person’s silhouette onto the principal axes x
and y. Examples of projection histograms are
depicted in Figure 5.
Figure 3. Human silhouette of the Foreground
image
Next from the silhouette, skeleton of
the human being is formed. It should be
obtained by using Thinning algorithm. It Figure 5. Projection histogram of human
reduces all black components in a binary silhouette
image to single pixel wide branches shown in Given the foreground mask F(x, y), the
figure 4 and preserves the following projection histograms Ɵ and π can be
properties: 1) Does not remove end points, 2) mathematically defined as:
does not break connectedness, and 3) does not
cause excessive erosion of the region. The
algorithm is described in detail in Gonzales
and Woods [8].
where the function Ø is equal to 1 if
F(x, y) is true, 0 otherwise, while Fx and Fy
are the width and the height of the foreground
mask F respectively.
In practice, O and . can be considered
as two feature vectors used to describe the
current frame.
V) Results and Discussions
Figure 4. Skeleton of the human silhouette
In order to evaluate the overall system
C) Projection Histogram performance, we apply the proposed approach
to a set of videos recorded in our lab. Here we
Since the videos were acquired by a consider indoor environment settings with dual
fixed camera, each frame I(x, y) is processed camera monitoring scenes. Distance of person
to extract the foreground mask (F) by means of to the camera is approximately 2 meters. We
a background subtraction step. could expect two different kinds of behaviour:
Normal, Unusual.
a) Normal Daily Activity of 11th International Conference on Computer
and Information Technology, 25-27, 2008.
Five different normal daily activities
are considered: Walking, Running, Bending [3] T. Lee and A. Mihailidis, “An intelligent
and rising up, Sitting down on the floor and emergency response system: preliminary
standing up and Lying down. development and testing of Automated fall
detection”, Journal of telemedicine and
b) Unusual -fall telecare, 194–198, 2005.
As most falls occur during intentional [4] Nait-Charif and S.J. McKenna, “Activity
movements initiated by the person, they Summarisation and fall Detection in a
happen mainly in forward or backward: Supportive Home Environment”, 17th
stumbling on an obstacle during walking, International Conference on Pattern
backward slip on wet ground. Recognition, 2004.
[5] Mohamad Hoseyn Sigari, Mahmood Fathy,
“Real-time Background Modelling/Subtraction
VI) Conclusions and Future Work using Two-Layer Codebook Model”,
Proceedings of the International
In this paper we developed a novel MultiConference of Engineers and Computer
real-time video surveillance system principally Scientist, 19-21 March, 2008
dedicated to fall detection. We claim that the
proposed system is not just an ordinary human [6] Rita Cucchiara, Costantino Grana, Andrea
fall detection system; it has many applicable Prati, and Roberto Vezzani, “Probabilistic
properties and can be employed in different Posture Classification for Human-Behaviour
surveillance systems. Moreover while existing Analysis IEEE Transactions on Systems, Man,
fall detection systems are only able to detect and Cybernetics”, Systems and Humans, Vol.
occurrence of fall behaviour, the proposed 35, Jan-2005
system is able to detect type of fall incident
(forward, backward or sideway). We used [7] Bruce Gooch, “Silhouette Extraction”,
combination of Horizontal and vertical Interactive non-photorealistic technical
projection histograms to detect the falls. illustration. Master’s thesis, University of
Finally A software program is developed to Utah, December-1998.
inform fall through mobile phone to care
takers. [8] Gonzales and Woods. “A thinning
algorithm is not the same as a medial axis
Future works will include the transform (MAT)”, page 491-494.
incorporation of multiple elderly monitoring
which is able to monitor more than one person
in the scene and also be able to handle
occlusion. Detecting additional behaviour is
also a subject to be explored in the future
work.
References
[1] Arie Hans Nasution and Sabu Emmanuel
Intelligent “Video Surveillance for Monitoring
Elderly in Home Environments”, IEEE
International conference, 2007.
[2] Homa Foroughi, Baharak Shakeri Aski,
and Hamidreza Pourreza “Intelligent Video
Surveillance for Monitoring Fall Detection of
Elderly in Home Environments”, Proceedings
