Smoke Alarm Detection, Broadcast Notifications and Social

Smoke Alarm Detection, Broadcast Notifications and Social
Smoke Alarm Detection, Broadcast
Notifications and Social Implications
Alan Woods
The Information Science
Discussion Paper Series
Number 2010/06
November 2010
ISSN 1177-455X
University of Otago
Department of Information Science
The Department of Information Science is one of seven departments that make up the
School of Business at the University of Otago. The department offers courses of study
leading to a major in Information Science within the BCom, BA and BSc degrees. In
addition to undergraduate teaching, the department is also strongly involved in postgraduate research programmes leading to MCom, MA, MSc and PhD degrees. Research projects in spatial information processing, connectionist-based information systems, software engineering and software development, information engineering and
database, software metrics, distributed information systems, multimedia information
systems and information systems security are particularly well supported.
The views expressed in this paper are not necessarily those of the department as a
whole. The accuracy of the information presented in this paper is the sole responsibility of the authors.
Copyright remains with the authors. Permission to copy for research or teaching purposes is granted on the condition that the authors and the Series are given due acknowledgment. Reproduction in any form for purposes other than research or teaching is forbidden unless prior written permission has been obtained from the authors.
This paper represents work to date and may not necessarily form the basis for the authors’ final conclusions relating to this topic. It is likely, however, that the paper will appear in some form in a journal or in conference proceedings in the near future. The authors would be pleased to receive correspondence in connection with any of the issues
raised in this paper, or for subsequent publication details. Please write directly to the
authors at the address provided below. (Details of final journal/conference publication
venues for these papers are also provided on the Department’s publications web pages: Any other correspondence concerning the Series should be sent to the DPS Coordinator.
Department of Information Science
University of Otago
P O Box 56
Fax: +64 3 479 8311
Smoke alarm detection, broadcast notifications
and social implications
Alan Woods
Information Science Department
University of Otago, Dunedin, New Zealand
Information Science Discussion Paper 2010/06
November 2010
Changes in population demographics and lifestyle choices have led to an increased risk
of higher mortality from house fires. The current average of 27 house fire related deaths
per year is likely to be exceeded in the following years. The aging population with its
natural increase in age related hearing loss and the younger demographic only having
mobile phones and no land-lines means there is a need for alternative warning methods
of smoke alarm activation. This project develops a proof of concept application that runs
on a smart phone and detects an activated smoke alarm. If there is no response by the
occupants automatically trigger an alarm to a predefined contact group. This application
can reduce the possibility of death or injury by persons unable to respond to an activated
The increasing availability of programmable mobile devices, smart phones in particular, creates
new opportunities. These devices can be used for various context dependent computation,
data gathering and tracking. Many smart phones offer today Internet and GSM connectivity
together with a range of sensors that can be used for improving quality of life, well-being and
There is an rapidly increasing proportion of the population that is at higher risk in the event
of a smoke alarm activation, this is due to a number of factors:
1. an aging population increasingly living alone [1],
2. age related hearing loss in the frequency range of the smoke alarms [2],
3. increases in no land-line connection to residences [3],
4. the anecdotal perception of the high cost of off site monitoring of alarms.
These factors combined have somewhat reduced the value of the work undertaken in the last
10 years to get smoke alarms into residences. Mobile smart phones are an increasing part of the
market. By developing a smoke alarm detection application that can be deployed on a mobile
device the risks to the population of adverse effects from fires can be reduced at little or no
cost to the public. The application provides a verification of actual fire situations to emergency
authorities by the independent verification process.
In New Zealand there is on average 27 house fire related deaths per year [1]. Age-specific
mortality rates were highest for adults ≥ 65 years and children < 5 years. Age-standardised
rates showed a threefold increased mortality for Maori compared with non-Maori [4]. While
the greater proportion of these deaths may have occurred in households that have no smoke
detection, some are due to the people not hearing the alarm. The use of a mobile device such
as a cell phone that can detect an activated smoke alarm and if there is no response by the
occupants automatically trigger an alarm to the emergency services could reduce the house fire
related mortality. The application strength is not only notification to the emergency services,
but also to neighbours and care-givers of a given person. We explore these aspects later in the
Environmental factors
New Zealand housing stock has a high proportion of dwellings that are built of wood [5]. The
housing is predominantly low density, this sets up situations where fires occurring in such
buildings are less likely to be noticed by other than the occupier and also will have a rapid
spread. Smoke detection systems are essential and important in the management of fire risk
in these homes. Currently the major type of smoke detector in New Zealand homes are the
battery powered ionising or optical smoke detectors. These detectors are inexpensive and are
able to be place anywhere. One of the shortcomings of these systems is that they have to
have their batteries replaced at regular intervals [6]. It is rare that this type of detector is
remotely monitored, and the system relies on occupant hearing the alarm and reporting the
event to emergency services. Most smoke alarms available in New Zealand resonate in the
range of 2 to 4 kHz. Coincidentally, it is also the frequency band most effected by age related
hearing loss[7]. Sprinkler based and mains powered smoke and CO2 systems are rarely installed,
possibly due to the relatively old housing stock and the high cost of installing and maintaining
these systems. Fire detection alarms are not mandatory in New Zealand homes although there
is a high proportion of the housing stock that has at least one alarm [4]. False alarms are an
issue with the basic smoke alarms used in households with up to four out office alarms proving
to be false, they can be set off with steam or smoke from cooking when no actual fire is present.
Human factors
Human factors are a major influence in the ability of the smoke alarm to warn occupants of
a potential fire. The primary one being the replacement of the batteries and testing of the
devices, there is also the issue of disabling the device due to multiple false alarms[8]. Response
to the alarm can be affected by a number of factors, such as: hard-of-hearing, children, older
adults, young adults, alcohol, mentally or drug impaired. New Zealand fire mortality statistics
show that the very young and the elderly are most at risk [4]. The young adult is a risk also
due to alcohol or drug impairment, and the naturally risky behaviour of the young adult [9].
Mobile devices
Many modern smart phones such as Android1 and Apple2 phones are equipped with microphones that can be used to detect sound in the local area. This detected sound can be processed
and analysed by the device and used to identify sounds. The devices also have built in GPS
units3 that can make the device aware of its geo-spatial location.
Smart phones have built-in accelerometer and GPS units and they act primarily as a communications device (that can be used in multiple different ways: SMS, phone call, or a data
connection). Modern smart phone devices are flexible enough to be re-purposed for other tasks
such as motor vehicle crash detection [10], portable ophthalmology examinations [11], and potentially many other health monitoring applications.
Principal Research Hypothesis
We propose that mobile devices can accurately discriminate smoke alarm warnings of standard
home smoke alarms available in New Zealand from the general day to day household sounds.
The discrimination will be to at least the 99.9th percentile (1 in 1000). This discrimination is to
the detection of the alarm not that there is a fire. Based on the fire alarm detection, the device
can autonomously, if not instructed otherwise by the owner, notify care givers, neighbours and
emergency services about the alarm.
Scope of the Research
Development of a proof of concept application on a mobile device, that will reliably detect the
majority of domestic smoke alarms in New Zealand and be able to give location aware warning
to emergency services in the event the user is unable to respond to the alarm.
Importance of the Study
With an increasing number of households not having land-line capability and the younger demographic predominately using cell phone technology[12] there is a need to provide automated
alarm connectivity for households without 3rd party alarm connections. These alarm connections are expensive and generally are outside the financial reach of the demographic most at
risk, those ≥ 65 and low decile households. Alarms currently available in the market today have
centre frequencies in the range of two to four kilohertz, this is the frequency range most effected
by age related hearing loss. The elderly are already over represented in house fire mortality
statistics[4]. New Zealand has a rapidly aging population, a population that increasingly lives
in single person dwellings[1]. This will lead to further increases in the fire fatality statistics for
this demographic.
Similar Studies
There appear to be no similar studies into the detection by mobile devices of smoke alarms.
There are however algorithms used to detect and identify songs, or sound. These use spectral
3 Global
Positioning System – a system used to locate the device.
fingerprinting of songs that is then compared to a central database [13]. Fingerprinting is a
common method of analysis of songs. These fingerprints are hash mapped [14] which provides
a fast access comparison path. While this is more complex than is required for detection of the
smoke alarms, the same principle will be used.
Design concepts
Smoke Alarms
The two most commonly recognised smoke detection technologies are ionisation smoke detection
and photoelectric smoke detection.
Ionisation-type smoke alarms have a small amount of radioactive material between two
electrically charged plates, which ionises the air and causes current to flow between the plates.
When smoke enters the chamber, it disrupts the flow of ions, thus reducing the flow of current
and activating the alarm.
Photoelectric smoke detection is generally more responsive to fires that begin with a long
period of smouldering. Photoelectric-type alarms aim a light source into a sensing chamber at
an angle away from the sensor. Smoke enters the chamber, reflecting light onto the light sensor;
triggering the alarm.
The alarm mechanism on all commonly available alarms use high output piezoelectric
buzzers. These buzzers have a defined fundamental frequency with even harmonic overtones.
The sound pressure output is in the range of 65 to 108 decibels (dB).
Alert systems
Direct communication with emergency services of an alarm while desirable, is not practical.
False alarms account for 90% of all alarm activations [15]. For this reason the alert system is
based on the neighbourhood watch concept instead. A group of neighbours would be listed in
the phone as contacts in the event of a fire. On an alarm activation the phone would group
text all the people in the fire contacts list. This can be extended to care givers and supervisors
of group homes etc. The people on the contact list can then ascertain the status of the person
who has activated the alarm and if needed, call the emergency services.
The use of SMS messaging is the most practical method of mass communication with the
contact list. A potential risk for this method of communication is that it has only best effort
quality of service and cannot be guaranteed to provide a timely service. Direct voice conference
calls are not easily implemented on mobile devices. While there are conference call facilities
on mobile devices they are not a free service and require to be set up by a telecommunication
For our proof-of-concept implementation we have used smart phone system equipped with
the Android operating system and application framework. Android is an open platform widely
used in a number of modern smart phones, tablets and some netbooks. Figure 1 demonstrates
the elements involve in the communication chain. A smoke alarm gives off an audible warning,
this warning is captured by the mobile phone application which analysis the sound to decide if
it is a true alarm. After deciding the alarm is valid, it activates a warning on the phone if the
user does not respond in 60 seconds it texts via Short Message Servicing (SMS) all the people
in the fire contact list.
Contacts Mobile Phones
Smoke Alarm
Mobile Phone
Cellular Network
Contact list
Figure 1: Communication Concept.
Application logic
To reduce the possibility of false positive alarms, when the analysed frequency transform of the
sound matches a alarm pattern in the database the incoming signal is sampled a further 10
times this reduces the chance that other background noise could have the same fundamental an
overtone frequencies. In normal music and day to day sounds it is rare to get exactly the same
frequencies occurring over time.
A 60 second delay is started when an alarm is detected this is to give time for the user to
react to the alarm and terminate the automatic sending of the text messages.
Figure 2: Flow diagram of alarm triggering.
Application architecture
Some elements of the application architecture are built into the Android phones, elements such
as contact lists and SMS messaging. The application links into these existing elements to
provide required services.
Figure 3: Design Concept.
Sound engine module
The heart of the application lies in the sound engine. This module sets up and reads in the
audio from the microphone. The module then sends the sample to the FFT for processing,
the returned FFT data is the used by the sound engine to ascertain the fundamental and
first overtones of the sample. The result is then compared against the existing database of
alarms and if there is a match is streamed through the audio buffer which is sampled and the
sample sent to the FFT module. After frequency transformations the resulting data s sent back
to the sound engine for further processing. The transformed data is filtered to establish the
fundamental and first overtone frequency of the sample. These values are then compared with
the database of alarms. If there is a match ten times in a row an alarm message is passed to
the sound module .
FFT module
This module takes the captures sound sample, performs a Fast Fourier Transform on it an
returns the result to the sound engine.
Sound module
This module provides a basic front end to the application (Figure 4a), it gets all the alarms
from the database (to reduce access times) and starts the sound engine module (Figure 4b).
It has a special menu system that allows for entering the new alarms and exiting the application(Figure 4c).
(a) At Startup
(b) Running
(c) Menu
Figure 4: Sound module front end .
Alarm module
This module is started when an alarm occurs, the module displays a flashing message on the
screen to warn an alarm is detected. If after 60 seconds the alarm cancel button has not been
pressed, the module looks up all the fire contacts in the contacts list and sends these to the
SMS module (Figure 5).
SMS send module
This API is part of the Android version 2.1. The alarm module sends the fire contact list to
the SMS module which then sends all the messages.
Add alarm module
The ’Add alarm module’ records the smoke alarm and saves it to an SQLlite database. There
are options to store the alarm type and the address location of the alarm to the data base as
well (Figure 6).
Add contacts module
The built in phone contacts module of the Android Version 2.1 operating system was used. A
contact type “Fire“ is made and the contact phone details added. This is then checked in an
alarm situation and used as a contact group list (Figure 7).
Figure 5: Alarm module display.
Figure 6: Add alarm module display.
Figure 7: Add contacts module display.
Base Frequencies
An analysis of smoke alarms supplied by the New Zealand Fire Service showed that all of
the alarms used high frequency piezoelectric crystals, ranging in fundamental frequency from
2953Hz to 3468Hz. Although there are Low frequency (under 500 Hz) alarms on the market
overseas there were none in the alarms supplied.
First Alert
First Alert
Fundamental Freq
1st Overtone Freq
Table 1: Alarm Fundamental and 1st Overtone frequencies
Minimum sound threshold
Testing Equipment:
• Sound level meter Radioshack cat no 33-2050.
• Noise generator: Toshiba M110 computer with Audacityr1.3.12-beta
Sound level meter Radioshack cat no 33-2050.
Test settings : sound pressure level (SPL), ’A’ weighted, fast response.
Test protocol:
• Phone and SPL meter 500 mm from sound source
• Measure ambient noise
• Start application
• Using 1 kHz tone start at ambient noise level, increase level till alarm triggers.
• Using 2 kHz tone start at ambient noise level, increase level till alarm trigger.
• Using 4 kHz tone start at ambient noise level, increase level till alarm trigger.
• Using 8 kHz tone start at ambient noise level, increase level till alarm trigger.
The application was tested to find the minimum sound threshold that would reliably trigger
the alarm.
Ambient noise level: 62db
Table 2: Alarm trigger threshold.
False triggering
Testing Equipment:
• Sound level meter Radioshack cat no 33-2050.
• Noise generator: Toshiba M110 computer with Audacityr1.3.12-beta
Test settings : sound pressure level (SPL), ’A’ weighted, fast response.
Test protocol:
• Phone and SPL meter 500 mm from sound source.
• Set volume to average threshold detection level 8.2.
• Start application .
• Play music tracks.
• Set volume to average threshold detection level + 20db 8.2.
• Play music tracks.
For the detector to be successful it should be immune to false triggering by sounds other
than alarms. For the purpose of this test a range of music was played at the average threshold
detection level and at 20db above threshold. Table 3 shows the range of music used and the
results of the testing.
Average threshold detection level: 65db
Slide Guitar
Death Metal
Nobody hold me close
To the Wall
Maple leaf rag
Black satin
Dave Hole
Chris Barber
Saint Sauns
Katie Webster
Table 3: False trigger test.
Alarm Triggered
Threshold +20db
Discrimination in noisy environments
It is important that the application is able to detect the smoke alarms when in noisy environments. The purpose of this test is to ascertain the minimum difference in sound levels between
background noise and the alarm sound to cause reliable triggering of the alarm.
For this test the same protocol as used in Section 8.3 with the ambient threshold replaced
with music. The average SPL values were set at 60db, 70db, 80db and 90db. The alarm tone
was triggered below the test threshold and the raised until the alarm triggered.
Background Noise (db)
Alarm Trigger value (db)
Difference (db)
Table 4: Alarm Trigger above background.
It can be seen from Table 4 that at all SPL levels there was little variation in the trigger
level above the background noise. This constant alarm trigger threshold value gives confidence
that the alarms are detectable in all high noise level environment as long as there is a minimum
of 7db difference in alarm SPL.
There is good discrimination between alarm activations and the surrounding noise, even in very
noisy situations there is no false positives. The detection algorithm is sensitive and will activate
the alarm at only 6db above the ambient noise level.
Continuous sampling of the microphone to ensure detection of all alarm events reduced the
battery life of the phone substantially. There was insufficient time to evaluate the current drain
of the application and this will need to be investigated further.
Continuous sampling for alarms was successful but at the cost of very high processor overhead, instead of continuous sampling it would significantly reduce the processing cost if a 5
second snapshot of detection occurred every 30 seconds. The interval between samples would
have to be carefully considered to trade off between improved processing and the increased risk
of greater fire involvement.
A further improvement in the processing costs could have been gained by the reduction of
the 16kHz sampling rate of the audio to 8 Khz , however, because of the Nyquist criteria of FFT
this would then reduce the detection to just the fundamental frequency as the first overtone
frequencies would be outside the FFT window. Future analysis should be done to see if it would
give acceptable results at this lower frequency.
The Android SMS API made it difficult to easily set up fire contact groups, for this reason
the fire contacts were set-up using google contacts that were then shared with the phone. The
latest API update has apparently rectified this problem.
Conclusion & Future work
We have successfully shown that detection of smoke alarms by mobile phone is possible, mobile
phones can be reliably used detect smoke alarms and warn owners and remote contacts of a
alarm situations. The alarms are detect reliably even in situations where there is high ambient
noise. A weakness of the system is in the processing overhead used to constantly monitor the
audio input of the phone. This application demonstrates the mobile phones ability to be used
as a advanced warning device in the event of a smoke alarm activation, its automated group
alert system reduces the chance of harm to people who have not been able to respond to an
Future Work
There are two major areas to be investigated further. The primary one is the implementation
of the sound engine and the FFT transform to run as stand alone background tasks. These
background tasks will call foreground task only when interaction with the user is required.
This will improve the overall architecture and design of the system. It should also improve the
battery consumption and overall responsiveness of the system.
Secondly, further research needs to be done on how often the input should be sampled and
if the sampling rate of the data can be lowered without reducing the reliability of the detection.
This has implication in the battery usage.
Further improvements to the Group Contacts module should be made using the latest API,
to allow adding both current address book contacts and new contacts to the ’Fire’ group in a
integrated user friendly interface.
A geolocation module was originally envisaged for this application, but problems with getting the GPS to work reliably inside multi-story dwellings and with dwellings that have metal
roofs meant there was insufficient time to get this feature working. The acquisition time of the
GPS is too long in the event of an alarm activation so location will have to be already cached on
the phone. Implementation of location awareness by use of GPS should be implemented. There
are constraints on the use of GPS inside dwellings due to signal level problems that need to be
overcome. Consideration of coarse location detection by cell tower triangulation combined with
cached GPS location data is a possible solution to the problem that should be investigated.
This application has only proved the proof of concept of the detection of smoke alarms using
the microphone input of mobile devices. The concept should be expanded to allow for input
from blue-tooth ,wireless devices and the current 434Mhz Rf links.
While using the application on a mobile device allows for portability, a fixed location device
that is mains powered would overcome many of the problems. A fixed location device should
be constructed and trialled in a number of test situations such as rural locations and single
person dwellings.
We would like to thank Senior Station Officer Barry Gibson from Southern Fire Region, New
Zealand Fire Service, for help with equipment testing and support during the course of this
We would like to thank Dr. Mariusz Nowostawski for supervising the research and helping with
the manuscript.
[1] New Zealand Department of Statistic, “1996 New Zealand Census.” Web, 1996.
[2] A. Lee, “The audibility of smoke alarms in residential homes,” US Consumer Product
Safety Commission, Bethesda, MD, 2005.
[3] S. Blumberg and J. Luke, “Wireless substitution: Early release of estimates from the national health interview survey, July-December 2008,” National Center for Health Statistics,
[4] M. Duncanson, P. Reid, J. Langley, and A. Woodward, “Unintentional house fire deaths
in new zealand 1991-1998.,” New Zealand Public Health Report, vol. 7, no. 7/8, pp. 31–40,
[6] M. Ahrens, N. F. P. A. F. Analysis, and R. Division, Smoke alarms in US home fires.
National Fire Protection Association, Fire Analysis and Research Division, 2009.
[7] T. I. Bruck D, “Smoke alarms for sleeping adults who are hard-of-hearing comparison of
auditory, visual, and tactile signals.,” Ear and hearing, vol. 1, no. 7, pp. 73–80, 2009.
[8] D. Bruck, “The who, what, where and why of waking to fire alarms: a review,” Fire Safety
Journal, vol. 36, no. 7, pp. 623–639, 2001.
[9] D. Bruck and M. Horasan, “Non-arousal and non-action of normal sleepers in response to
a smoke detector alarm,” Fire Safety Journal, vol. 25, no. 2, pp. 125–139, 1995.
[10] C. Thompson, J. White, B. Dougherty, A. Albright, and D. Schmidt, “Using Smartphones
and Wireless Mobile Sensor Networks to Detect Car Accidents and Provide Situational
Awareness to Emergency Responders,”
[11] V. Shah, R. Lord, and A. San Filippo, “Smartphones in ophthalmology: Potential uses of
the smartphone in your practice,” Feb 20 2010.
[12] S. J. Blumberg and J. V. Luke, “Wireless substitution: Early release of estimates from the
national health interview survey.,” National Center for Health Statistics, May 2009.
[13] A. Ramalingam and S. Krishnan, “Gaussian mixture modeling of short-time Fourier transform features for audio fingerprinting,” Information Forensics and Security, IEEE Transactions on, vol. 1, no. 4, pp. 457–463, 2006.
[14] D. Conklin, “Discovery of distinctive patterns in music,” Intelligent Data Analysis, vol. 14,
no. 5, pp. 547–554, 2010.
[15] M. Ahrens, “US experience with smoke alarms and other fire detection/alarm equipment,”
National Fire Protection Association, 2007.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF