Conference PaperPDF Available

Application of Sensor Networks in a Smart Apartment

Omid Salehi-Abari*, Furkan Alaca*, James R. Green, and Rafik Goubran
Department of Systems and Computer Engineering, Carleton University
*O. Salehi-Abari and F. Alaca contributed equally to this work
Home-based health monitoring systems provide
an opportunity to enhance the quality of life for older
adults by allowing them to live independently and
improving doctor-patient efficiency. These systems are
more cost effective than constant medical supervision
in an institutional setting and will help in reducing the
load on the healthcare system [1]. Studies show that
people aged 75 and older are significantly more likely
to require health care services than other adults [2].
In this paper we propose a number of non-invasive
applications of low-cost and low-powered wireless
sensor networks aimed to monitor various aspects of a
user’s health, daily activities, and environment. Our
goal is to increase the perceived level of privacy by the
subject as compared to more invasive monitoring
technologies such as video cameras. Our approach is
to place the sensors in an unobtrusive fashion and to
avoid using imaging sensors.
Measurements from simple temperature, humidity,
and light sensors can be combined to ultimately aid in
health monitoring and detecting medical emergencies.
A number of other initiatives have used diverse
sensors to track human health and emergencies (e.g.
[3], [4], [5]). In the present study, we investigate the
use of low cost, low power wireless sensors for
tracking indicators of human health. As discussed
below, we have actually implemented these systems
and conducted a number of proof-of-concept
We propose algorithms which can be used to help
detect shower usage, monitor exposure to natural
sunlight, and collect data on day-to-day activities such
as refrigerator and bathroom use. The information
collected can then be forwarded to health care
professionals or emergency contacts. This work was
completed during an 8-month senior design project.
Although this is a pilot study, we aim to deploy the final
sensor network and data fusion system in an
instrumented Smart Apartment as part of the TAFETA
initiative [1].
Wireless sensor hardware
In this study, we have employed the Crossbow Iris
mote. The mote uses an IEEE 802.15.4 compliant RF
transceiver and uses the unlicensed 2.4GHz band.
The mote has a maximum indoor range of 50m, which
is appropriate for a one-bedroom apartment. One of
the motes was configured as a base station by
connecting it to a desktop computer via a USB
connection and loading the appropriate software. The
other motes can then wirelessly transmit data to the
base station using the specified sampling rate. The
base station automatically saves the data from the
sensors into an SQL database. This allows us to
analyze the data in real-time. MATLAB was used to
retrieve the data from the SQL database and to
perform the required operations on the data.
Overview of application areas
We explored a number of possible applications
where the sensors can be used either independently
or in conjunction with other instrumentation. Figure 1
shows the floor plan of the one bedroom apartment
used by TAFETA and our proposed sensor locations.
Figure 1: Proposed sensor locations in the one
bedroom apartment
Toilet usage
Toilet usage can be monitored by placing a
temperature sensor inside the tank, as shown in
Figure 2. Since flushing the toilet expels the tepid
water from the tank and replaces it with colder water
from the pipes, there will be a measureable drop in
temperature. Therefore, by collecting periodic
temperature measurements and applying a
differentiator to the raw data, we can detect toilet
flushing events.
Figure 2: Photo of temperature sensor in toilet tank
Shower usage
Shower usage can be detected by monitoring the
humidity in the bathroom. A significant increase in
humidity typically indicates that the shower is in use.
Shower detection can be helpful for two purposes.
Firstly, keeping a record of shower usage can help
ensure that the subject is maintaining their personal
hygiene activities. Secondly, abnormally long showers
can be detected if the humidity remains high for a
prolonged period. Before this can be accomplished,
sufficient training data must be collected to establish
the average time which the person typically takes in
the shower. An excessively long shower could indicate
that the person requires assistance.
Refrigerator usage
Refrigerator usage can be monitored by using a
light sensor inside of the refrigerator. When the
refrigerator door is opened, there will be a significant
increase in light intensity. Since the light intensity
reading will be zero when the fridge door is closed,
even the ambient light from the room will be enough to
detect an open door. Using a light sensor could have
advantages over other methods, such as magnetic
switches, since it may be easier to install and it does
not change the external appearance of the refrigerator.
The light sensor also proved to be more accurate than
the temperature sensors as discussed below.
Monitoring the refrigerator door can give an indication
of whether or not the subject is following their regular
eating pattern. In addition, it can help to identify cases
where the door is accidentally left open. Note that the
refrigerator usage could perhaps be more simply
monitored by using a contact sensor in the door. The
current solution is preferred for two reasons: 1) we are
using the same commercially available wireless sensor
mote for all applications and have avoided the need for
specialized hardware for each monitoring application;
2) our solution is suitable for retrofitting on any
refrigerator and does not require any modification of,
nor integration with, the appliance.
Exposure to natural light
It has been shown in studies that older adults are
more likely than others to have vitamin D deficiencies
due to lack of exposure to natural light [6]. Therefore, it
is of benefit to track a subject’s exposure over time.
Light measurements can be taken throughout the
day to determine the subject’s exposure to natural
light. To differentiate between artificial and natural light
sources, temperature sensors can be placed adjacent
to lamps, which can determine whether the lamp is on
or off based on the heat relative to room temperature.
By combining light and temperature data, we can
detect whether the room is lit by artificial or natural
light. Also, if the subject leaves their lights on during
the day, they can be advised to open their curtains.
Room temperature measurements
The sensor network could help determine whether
a window is open or closed. For example, temperature
measured near the window could be compared with
the aggregate temperature from the sensor network. In
an experiment involving three sensors, the state of the
window (open/closed) was clearly discernable from the
temperature data. While this approach will only work
when a significant temperature differential exists
between indoors and outdoors, this is precisely when
window status information is most useful.
The sensor network can also help determine
whether this is an appropriate time to open the
window. Temperature and humidity data could be
collected from both inside and outside. This
information could then be compared in order to inform
the subject of whether or not it is an appropriate time
to open the window for fresh air. This could be of
particular benefit to patients suffering from reduced
mental acuity or dementia. A number of proof-of-
concept experiments were successfully conducted to
demonstrate the feasibility of this approach.
Data was collected and analyzed for each of the
aforementioned applications. The sampling rate for
each mote was chosen based on the application and
the type of sensor (1min for toilet, 1min for shower,
15s for refrigerator, 5min for natural light). For all
experiments, the ambient room temperature was 25ºC.
Different types of data can change at different rates
depending on their time correlation. For example, light
intensity can change almost instantaneously, whereas
changes in temperature and humidity are more
gradual. Consequently, using temperature sensors
allows us to use a lower sampling rate without missing
any events. Lower sampling rates result in fewer
transmissions, which allows for lower power
consumption and longer battery life.
Toilet usage
Figure 3 shows the water temperature in a toilet
tank. As it can be seen, the water temperature falls
suddenly when the toilet is flushed. This permits the
detection of flush events by using a differentiator and a
threshold algorithm. Many hours of data were collected
in order to test the algorithm. As a result we were
successfully able to detect toilet flushes which were at
least 20 minutes apart. This value is reasonable
assuming that only one person is living in the
apartment. Toilet flushes which were closer than 20
minutes were undetectable due to the slow recovery of
the tank water to room temperature.
Figure 3 : Water temperature in toilet tank (flushes are
indicated by arrows)
Shower usage
In order to detect the shower usage, we placed the
humidity sensor 1m away from the shower. We found
that the humidity rose sharply after approximately one
minute. However the drop in humidity had a large
delay after the completion of the shower. These delays
depend on factors such as bathroom size and sensor
location. The state of the exhaust fan (i.e. on/off) also
impacts the rate of decrease in relative humidity
following a shower (but does not delay the increase in
humidity at the start of at the shower). The shower
detection algorithm worked regardless of the state of
the fan. Although we were not able to detect the exact
shower duration, we were still able to detect the
shower usage event. In addition, it may be possible to
detect excessively long showers provided that delays
are constant for a given installation. An excessively
long shower could be an indication that the subject
requires assistance.
Refrigerator usage
Figure 4 shows the light intensity measured by a
sensor placed inside a refrigerator door. The peaks in
the plot indicate the events where the refrigerator door
was opened. The sampling rate for this application
needs to be relatively high in order to capture all of the
events. This is due to the abrupt nature of the changes
in light intensity. Initially, we tried to detect these
events by measuring the change in temperature in the
refrigerator door when it was opened. However, this
was found to be impractical due to the insignificant
change in temperature when compared to the
temperature oscillations caused by the refrigerator
compressor turning on and off.
Figure 4: Refrigerator light (light is zero when
refrigerator door is closed)
Exposure to natural light
Figure 5 shows the light intensity and the lamp
temperature during night hours. When the lamp is
turned on, there is a sharp increase in both the light
and temperature readings. For low sampling rates,
which allows for better battery life, we found that it is
more favourable to monitor the lamp using
temperature measurements. Since changes in
temperature are more gradual with respect to time, it is
more reliable to detect the events based on
temperature rather than light. As can be seen from
Figure 5, relying solely on the light sensor may lead to
undetected events if a low sampling rate is used. By
combining the light and temperature data, we will be
able to monitor whether the light is on or off throughout
Light (Lux)
the day. When the light is on, we will see a rise in both
temperature and light measurements. On the other
hand, if the subject opens or closes the curtains, a
sudden change will result in the light intensity graph
without any observable change in the temperature.
This is illustrated in Figure 6, which shows the light
intensity of the room and the lamp temperature during
daytime and evening hours.
Figure 5: Room light and temperature at night (Arrow
shows undetected event using light sensor; sampling
rate of 5 minutes used here.)
Figure 6: Room light and temperature during day-time
hours. (Closing and opening of curtains are indicated
by arrows. The unmarked peak is due to an artificial
light source.)
The system presented in this paper is designed to
monitor the health and day-to-day activities of the
subject. By using the temperature and humidity
sensors, we were able to monitor room temperature as
well as toilet and shower usage. Adding the light
sensors to the system allows us to monitor refrigerator
usage, room lighting, and the subject’s exposure to
natural light. The algorithms used in this study have
relatively low complexity. While more complex pattern
classification algorithms such as neural networks or
decision forests may be investigated in the future, the
low complexity of the current approach increases the
likelihood that the results will generalize to new
patients and new locales [7].
Future work will focus on integrating this system
with other smart home technologies. For example, light
measurements can be combined with bed occupancy
to monitor sleeping patterns (which is of great interest
for patients with senility). Lastly, while the Crossbow
mote hardware used here is convenient for rapid
prototyping, ultimately we plan to develop custom
hardware with only the subset of sensors required for
this specific application. For example, complex
sensors such as GPS are supported by the Crossbow
mote platform but are not required in the Smart
Apartment, and therefore a simpler microcontroller
may be selected for the final system. Through
optimization of parameters such as sampling rate,
choice of microcontroller, and choice of wireless
communication frequency band, we hope to achieve
lower cost, smaller size, and lower power
This study was supported by a grant from the
Natural Sciences and Engineering Research Council
of Canada.
[1] A. Arcelus, M. H. Jones, R. Goubran, F. Knoefel, “Integration of
Smart Home technologies in a health monitoring system for the
elderly", IEEE First International Workshop on Smart Homes
for Tele-Health, Niagara Falls, Canada,2007
[2] M. W. Rosenberg, E. G. Moore, ”The health of Canada's
elderly population: current status and future implications”,
Canadian Medical Association Journal, pp. 1025-1032, 1997.
[3] R. Aipperspach, E. Cohen, and J. Canny, "Modeling human
behavior from simple sensors in the home," Lecture Notes in
Computer Science, vol. 3968, p. 337, 2006.
[4] J. Chen, A. Kam, J. Zhang, N. Liu, and L. Shue, "Bathroom
activity monitoring based on sound," Third International
Conference, PERVASIVE, Springer, p. 4761, 2005.
[5] I. Neild, D.J. Heatley, R.S. Kalawsky, and P.A. Bowman,
"Sensor Networks for Continuous Health Monitoring," BT
Technology Journal, vol. 22, 2004, pp. 130-139.
[6] F. M. Gloth, C. M. Gundberg, B. W. Hollis, J. G. Haddad, J. D.
Tobin, Vitamin D deficiency in homebound elderly persons”,
JAMA, vol. 274, pp.16831686, 1995.
[7] R. O. Duda, , P.E. Hart, D.G. Stork, Pattern Classification, 2nd
ed., Wiley, 2001.
Light (lux)
Light (lux)
Internet of Things (IoT) continues proliferation. Hospitals, energy industry, power grid, food and water, transportation, etc. are critical asset resources. Critical systems must remain steadfast and reliable. This IoT evolution, sometimes termed“, IIoT” or Industrial IoT, is significant for critical infrastructure. In harsh climates, human lives can depend daily on these systems functioning well. Better situational awareness can improve resilience of these critical infrastructures and make them increasingly robust. In this paper, the method of Agile IoT for critical infrastructure resilience is presented, and an Agile IoT model is presented to use signal processing middleware to enable existing sensors commonly deployed in critical infrastructure to be agilely re-purposed to 1) add parallel communication methods and/or 2) sense additional mechanical / physical parameters using existing hardware. The results of the proposed method are applied to use common temperature sensors to detect non-temperature parameters: fluid flow in a pipe (manufacturing or water), ice buildup (transportation or energy infrastructure), and of mechanical door state (health venues of life-saving medicine stored in hospital refrigerators).
Full-text available
Continuous monitoring of well being is one of a number of new schemes under development that promise to reduce the UK's health care bill and ease the burden on carers and support organisations. Fundamental to its widespread adoption is the deployment of low-cost sensor networks in residential, sheltered and community settings. These networks must be capable of a variety of monitoring functions, the data from which is collated locally and then delivered to health care professionals over the national network. Delivering this data reliably and securely is central to BT's service offering. This paper describes our work to date in developing a sensor network that will be deployed in later trials of continuous well-being monitoring.
Conference Paper
Full-text available
Pervasive sensors in the home have a variety of applications including energy minimization, activity monitoring for elders, and tutors for household tasks such as cooking. Many of the common sensors today are binary, e.g. IR motion sensors, door close sensors, and floor pressure pads. Predicting user behavior is one of the key enablers for applications. While we consider smart home data here, the general problem is one of predicting discrete human actions. Drawing on Activity Theory, the Language-as-Action principle, and Speech understanding research, we argue that smoothed n-grams are very appropriate for this task. We built such a model and applied it to data gathered from 3 smart home installations. The data showed a classic Zipf or power-law distribution, similar to speech and language. We found that the predictive accuracy of the n-gram model ranges from 51% to 39%, which is significantly above the baseline for the deployments of 16, 76 and 70 sensors. While we cannot directly compare this result with other work (lack of shared data), by examination of high entropy zones in the datasets (e.g. the kitchen triangle) we argue that accuracies around 50% are best possible for this task.
Conference Paper
Among older adults, the challenges of maintaining mobility and cognitive function make it increasingly difficult to remain living alone independently. As a result, many older adults are forced to seek residence in costly clinical institutions where they can receive constant medical supervision. A home-based automated system that monitors their health and well- being while remaining unobtrusive would provide them with a more comfortable and independent lifestyle, as well as more affordable care. This paper presents a smart home system for the elderly, developed by the Technology Assisted Friendly Environment for the Third Age (TAFETA) group. It introduces the sensor technologies integrated in the system and develops a framework for the processing and communication of the extracted information. It also considers the acceptability and implications of this technology from the perspective of the potential occupants.
Conference Paper
In this paper an automated bathroom activity monitoring system based on acoustics is described. The system is designed to recognize and classify major activities occurring within a bathroom based on sound. Carefully designed HMM parameters using MFCC features are used for accurate and robust bathroom sound event classification. Experiments to validate the utility of the system were performed firstly in a constrained setting as a proof-of-concept and later in an actual trial involving real people using their bathroom in the normal course of their daily lives. Preliminary results are encouraging with the accuracy rate for most sound categories being above 84%. We sincerely believe that the system contributes towards increased understanding of personal hygiene behavioral problems that significantly affect both informal care-giving and clinical care of dementia patients.
To assess the vitamin D status in homebound, community-dwelling elderly persons; sunlight-deprived elderly nursing home residents; and healthy, ambulatory elderly persons. A cohort analytic study. Of 244 subjects at least 65 years old, 116 subjects (85 women and 31 men) had been confined indoors for at least 6 months, either in private dwellings in the community (the Hopkins Elder Housecall Program) or in a teaching nursing home (The Johns Hopkins Geriatrics Center). The 128 control subjects, a healthy ambulatory group, came from the Baltimore Longitudinal Study on Aging. All subjects were free of diseases or medications that might interfere with their vitamin D status. Serum levels of 25-hydroxyvitamin D (25-OHD) and 1,25-dihydroxyvitamin D (1,25-[OH]2D) were measured in all subjects. In a subgroup of 80 subjects, serum levels of intact parathyroid hormone (PTH), ionized calcium, and osteocalcin and intake of vitamin D (through 3-day food records) were assessed. A randomly selected cohort of sunlight-deprived subjects also had serum levels of vitamin D binding protein measured. In sunlight-deprived subjects overall, the mean 25-OHD level was 30 nmol/L (12 ng/mL) (range, < 10 to 77 nmol/L [< 4 to 31 ng/mL]) and the mean 1,25-(OH)2D level was 52 pmol/L (20 pg/mL) (range, 18 to 122 pmol/L [7 to 47 pg/mL]). In the sunlight-deprived subjects, 54% of community dwellers and 38% of nursing home residents had serum levels of 25-OHD below 25 nmol/L (10 ng/mL) (normal range, 25 to 137 nmol/L [10 to 55 ng/mL]). A significant inverse relationship existed between 25-OHD (ie, Log [25-OHD]) and PTH when they were analyzed together (r = -0.42; R2 = 0.18; P < .001) and for each cohort separately. All other parameters measured, except ionized calcium, differed significantly from the Baltimore Longitudinal Study Group means. The mean (SD) daily intakes of vitamin D (121 [132] IU) and calcium (583 [322] mg) were below the recommended dietary allowance only in the community-dwelling homebound population. The mean vitamin D binding protein level in the sunlight-deprived subgroup was in the normal range. Despite a relatively high degree of vitamin supplementation in the United States, homebound elderly persons are likely to suffer from vitamin D deficiency.
The growing size of Canada's elderly population and its use of health care services has generated much discussion in policy circles and the popular press. With data from the National Population Health Survey, undertaken in 1994-95, the authors examine the health status of Canada's elderly population using 3 sets of measures: level of activity limitations, prevalence of chronic illnesses and self-assessment of overall health. They also analyse the utilization of physician and institutional services. The profile of this population the authors develop is in many respects not much different from that of the remaining adult population, until the age of 75. People aged 75 and over are much more likely than other adults to have health problems and use health care services. Also, elderly women living alone and with low income are identified as an especially vulnerable group who need access to medical and nonmedical services if they are to remain in the community. Using Statistics Canada projection data the authors discuss some aspects of the elderly population's health status in the future. Their look into the future raises issues about the preparedness of health care providers and our health care system to meet the challenges of tomorrow's elderly population.
Modeling human behavior from simple sensors in the home
  • R Aipperspach
  • E Cohen
  • J Canny
R. Aipperspach, E. Cohen, and J. Canny, "Modeling human behavior from simple sensors in the home," Lecture Notes in Computer Science, vol. 3968, p. 337, 2006.
Vitamin D deficiency in homebound elderly persons
  • F M Gloth
  • C M Gundberg
  • B W Hollis
  • J G Haddad
  • J D Tobin
F. M. Gloth, C. M. Gundberg, B. W. Hollis, J. G. Haddad, J. D. Tobin, "Vitamin D deficiency in homebound elderly persons", JAMA, vol. 274, pp.1683-1686, 1995.