issue: August 2007 APPLIANCE Magazine
Ultra-Low-Resolution Thermal Imaging for Detecting Kitchen Hazards
Email this Article
by Justin A. T. Halls, Brunel Institute for Bioengineering, Brunel University, UK
More than 50% of fires are the result of unattended cooking appliances. This article describes a device that uses a very-low-resolution thermopile array to scan the kitchen to detect the presence of heat sources.
The kitchen is one of the most dangerous locations in the home. Fifty percent of reported fires and 66% of unreported fires are the result of cooking, and the majority of these are caused by leaving cooking unattended.1,2 This is especially a problem for elderly users whose short-term memory may be deteriorating, and frequently leads to additional concern by the elderly that they may have left the cooker on, or have left “something on the stove.”
A device that could detect that the cooker had been left switched on, whether or not there was food being cooked, and could determine that the cooker was unattended, would be useful for alerting the occupant to the risk, reducing the hazard, and providing reassurance to the user. The device should be of low cost, simple to install, and should be capable of either operating in a stand-alone mode or integrated into a more complex “caring home” system.3
This article considers in particular the task of sensing that the cooker is on. The task of alerting the user may be implemented as a simple audible alert for a completely stand-alone device, or the output may be linked into a smart or caring home system, which would then be responsible for alerting the occupant.
Several approaches are possible, including monitoring of the cooker supply services (usually gas or electricity); temperature sensors attached directly to the cooker, or located immediately above it; infra-red flame detectors, tuned to specifically detect CO2 emissions; and remote thermal-imaging systems.
It is especially important to keep the installation costs as low as possible. Therefore, any device that involves connecting directly to the gas or electricity supply is less desirable. It is also quite difficult to detect that gas is being used, and this would eliminate a considerable number of potential users.
Attaching a sensor directly to, or in the immediate vicinity of, the cooker is quite labor-intensive and also requires the device to be specifically tailored to each individual installation.
Remote infrared sensing, which could be implemented as a discrete module mounted on the ceiling some distance away from the cooker, would allow for simple installation. If located near a wall, it could be mains-powered, eliminating the need to change batteries regularly,
as well as providing a communication channel to a smart home host.
CO2-detecting flame detectors were rejected since the goal is to work with electric as well as gas cookers. However, conventional thermal imaging solutions, while very versatile, are extremely expensive.4
A simple, low-cost, eight-element linear array can provide quite adequate resolution if scanned slowly across the kitchen. It is not necessary to know which part of the cooker is in use, simply that some part of it is hot. A separate passive infrared (PIR) detector is able to determine whether the occupant is present and if an alarm needs to be sounded. For this prototype, the alarm rules were arbitrarily set to generate an audible alarm when the occupant left the kitchen while the cooker was on, and then to repeat the alarm if the kitchen is left unattended for more than 20 minutes. Thereafter, the alarm was repeated every 2 minutes. until the occupant returned.
Materials and Methods
Hardware. The device is based on a TSE 01/08 L linear eight-element thermopile sensor from HLPlanar Technik GmbH. The sensor is supplied on a small printed circuit board (PCB) with a built-in preamplifier and signal conditioner and 8:1 multiplexer. The sensor has a field of view approximately 4° by 40°.
In order to provide adequate coverage of the kitchen, the sensor needs to be scanned. In order to avoid the complication of reciprocating scanning or the use of power and data slip rings, the device provides scanning by rotating a polished stainless-steel mirror (see Figure 1).
Figure 1. Scanning head and sensor.
The mirror is rotated by a simple quartz electric clock mechanism that can be independently battery powered or can take its power from the main processor board. The drive is geared down from the second hand shaft to provide a complete revolution, which provides two image scans, every four minutes. The mirror moves in discrete steps of 1.5° every second, and a synchronization pulse is derived from the clock mechanism. An additional trigger pulse is derived from the gear mechanism to initiate each image scan.
Provision is also made for attaching a small laser pointer with its beam parallel to the axis of the sensor field of view. This pointer can be activated remotely. It can be used for alignment and could also be potentially used to highlight the area responsible for generating an alarm.
Most testing was done with the sensor mounted about 220 cm above floor level and with the axis of the mirror tilted approximately 30° from the vertical. This placed the center of the image at the center of the cooker hob.
Data collection and image processing were performed by a Microchip PIC18LF458, which is a very compact microprocessor incorporating a wide range of input and output facilities. Although not required for an installed system, during testing all data were recorded to a 2-MB MMC flash memory card. This allowed data to be downloaded to an Excel spreadsheet for more-detailed analysis and visualization.
An SGM-5910 PIR sensor from Nippon Ceramic Company, Ltd., was located adjacent to the scanner head to detect when the occupant was present in the kitchen. The output was stretched to ensure that any movement during a scan would be recorded.
An audible beeper was connected to the alarm output for test purposes. In normal use this could be a “pinger,” such as is often used in theaters to announce the end of the interval, or a connection to a smart home system that could deal with the alarm in an appropriate way.
Algorithms. The basic principles of operation are to identify the presence of hotspots within the image area. A hotspot is defined as an area that is hotter than a threshold value derived from the average of all the pixels in the image and which is not cooling down. The latter requirement is important because electric kettles and electric cooker rings may take up to 45 minutes to cool to a level where they are no longer distinguishable from the background.
If the PIR sensor changes from being active during one scan to being inactive during the next and there is a hotspot present, then the alarm is sounded. If the PIR has been inactive for 10 scans (20 minutes) and there is a hotspot present, then the alarm is sounded.
Data collection is triggered by the start of scan signal, followed by a predetermined number of tick triggers to allow the mirror to rotate to the desired starting point. Thereafter, a line of data is read following each tick trigger, with a 100-millisecond delay to allow the movement to settle. A complete image would comprise 90, 100, or 120 lines of eight pixels. Since it is only necessary to identify the presence of hotspots, only the hottest pixel in each vertical line was recorded. Thus, one image would consist of a single-dimensional array of 8-bit values.
Hotspots are identified as contiguous groups of pixels that are above a threshold value derived from the average of all pixels in the image. Data about the five most significant hotspots are retained, significance being derived from the sum of all the values in the hotspot. If a hotspot disappears, the data are retained unless the space is needed to store data about a more significant or a currently visible hotspot. This allows a hotspot to be temporarily obscured without being forgotten about.
Since many hot items remain visibly hot even after they have been turned off, it is important to determine whether a hotspot is cooling down. The stored history of a hotspot is used to identify whether the temperature of the hotspot is tending to cool down. If the trend is negative, the information about the hotspot is retained, but the hotspot will not trigger an alarm.
The scanner was installed in a small kitchen in a demonstration flat, which was equipped with an electric cooker, kettle, and fridge. The scanner was positioned near the ceiling above the sink and was tilted downward so that the scan would sweep along the length of the food preparation and cooking areas. Because of this tilt, the scanner ended the sweep pointing largely upward, and this allowed the fluorescent light in the kitchen to be included in the scanned area.
Figure 2. Time-temperature plot showing detection of an electric kettle and an electric cooker ring: (a) contour plot of entire sequence, (b) profi les at azimuth angles 45° and 56°.
Figure 2a provides a visualization of data collected over a period of just under two hours. Time runs from the bottom of the picture to the top, and the scan sweeps from left to right with 90 samples per scan.
At position 45, the kettle is detected when it is turned on at minute 12, and it rapidly comes to the boil and automatically switches off. At minute 40, a pan of water is put on to boil, and shortly after that, it is turned off and allowed to cool down on the cooker.
There are two other significant heat sources in the detection area. At position 75, there is a heat source that comes and goes in a regular pattern. This was an electric radiator that was visible through the kitchen door, cycling on its thermostat. Further over is a larger heat source that disappears after minute 60. Because the mirror axis was tilted, the scan was tilted up toward the ceiling at each end of the sweep. This heat source is due to the end of the fluorescent light fitting being in the field of view, and it disappears when the light is switched off.
Figure 2b shows individual profiles along the time axis. The red line indicates the highest value in the region around azimuth angle of 45°, and the blue line indicates the peak values in the region around azimuth angle 56°. These positions correspond to the location of the kettle and the cooker hob. For both areas, there is a rapid rise in temperature when they are turned on, followed by a long cooling period after they have been turned off. There is a local disturbance in these trends when the occupant is present in the kitchen for a while after turning the ring off.
Several areas that are cooler than the general ambient background are apparent in Figure 2a just after the pan has come to the boil. These are due to the presence of steam, which is a very efficient absorber of infrared. Because the background becomes invisible, the sensor sees this as a cold region.
Since we are only interested in determining whether cooking is in progress, not the actual temperature profile of the cooker, the presence of steam can be used to our advantage. The hotspot detection algorithms were modified to detect areas that were apparently colder than a threshold value derived from a long-term average of all pixels in the image. The presence of a cold area is then treated in the same way as the presence of an active hotspot by the alarm generation algorithm. When heat is removed from the source of the steam, the steam disperses very rapidly, and false alarms will not be generated by the presence of “old” steam.
Figure 3. Time course of a single hotspot, showing when a person is detected as being present (tall bars) and when the alarm is sounded (arrows).
Figure 3 shows a different experiment in which the extraneous heat sources have been excluded from the field of view, and also incorporates data from the PIR detector and the alarm output. The timelines displayed are those of hotspots caused by a kettle and a cooker ring with a pan of water. The occupant is present at minute 8 and turns the kettle on. The kettle then comes to the boil, switches itself off, and starts to cool down. At minute 24, the occupant returns and turns the ring on at full, bringing the water to the boil over the next 12 minutes. When the occupant leaves the kitchen, there is no initial ping since the pan has not yet warmed to the point of being recognized as a hotspot, and the kettle is recognized as cooling down. By minute 45, a cloud of steam has developed, obscuring the heat sources and showing it as a large cool area. This causes the boiling flag to be set, which is equivalent to detecting a hotspot, and since the occupant has been absent for 20 minutes, the warning ping sounds at two-minute intervals until someone returns and turns the cooker off. The warning sounds once more when the person leaves the kitchen, since there is still a hazard present, but the steam disperses and the ring cools down so that no further warnings are generated.
This shows that the system is capable of detecting hazards either from temperature, or because they are producing steam, and can distinguish between active heat sources and those that are turned off and cooling down. The system was also found to be capable of detecting the oven and grill as heat sources, although the threshold for detecting the oven will depend on the quality of the insulation.
Leaving cooking unattended is not only hazardous but is extremely worrying, especially for elderly users that may suffer from memory loss. Fire and smoke detection systems do not address this problem, as they are not triggered until a fire has occurred, rather than attempting to prevent the fire in the first place. Triggering of a smoke alarm can also be very disconcerting for elderly people and may cause additional injuries as they rush to correct the situation.
The current system has been shown to be able to detect normal kitchen heat sources very adequately. Even low-level heat sources, such as a moderately well-insulated oven or a heat ring left on its lowest setting, can generally be detected, and the system automatically compensates for changes in the ambient temperature.
Positioning and alignment of the device are important considerations. The device used in these experiments was limited in its range of adjustment, and the scanned area tilted up at the ends allowed it to detect the heat from light fittings. The preferred orientation is with the mirror axis kept vertical and the sensor position adjusted to align the field of view correctly. It would be desirable to be able to adjust the areas within the field of view that were important and to exclude areas that might contain “normal” heat sources such as a heater. For example, a row of miniature switches could be used to disable certain segments of the scan.
The distance of the sensor from the cooker is not critical. For these experiments, the area covered by a single pixel was about 30 cm across, but performance is not significantly degraded if only one or two pixels see the entire top surface of the cooker.
Alignment is simplified by attaching a small laser pointer to the device, coaxially with the line of sight of the sensor, and this shows the area being covered by the system. It is also possible to have the microcontroller switch the laser pointer on and off in order to highlight the areas where heat hazards were detected.
The way in which the warnings are generated is important for the device to be acceptable. Discrete pings can be quite penetrating but need not cause embarrassment if there are visitors present. A more complex speech-based warning could be possible using X10 protocols to trigger speech devices connected to local loudspeakers throughout the house, but is more intrusive. More-complex home warning systems, such as providing verbal warnings by means of a telephone call, could also be possible.5 It would also be desirable to link the device to a warning light located adjacent to the front door in order to alert occupants if they try to go out while the cooker is on.
The physical construction of the system needs some consideration. The kitchen is an inherently dirty environment, and the sensor and mirror surfaces would need to have some protection against dirt and grease. Also, if the device was battery powered, further refinement of the system to minimize power usage would be advisable.
The system has so far only been tested using an electric cooker with solid hotplates. It would be necessary to do further testing using a wide range of cooker types, including gas cookers. However, simple tests with similar sensors suggest that the system would work equally well with any cooker type.
- Fire Facts and Statistics, National Community Fire Safety Centre Toolbox, Office of the Deputy Prime Minister (January 2004).
- DTI HASS/LASS database, [online] (Edgbaston, Birmingham, UK: Royal Society for the Prevention of Accidents [cited 19 July 2007]); available from Internet: www.hassandlass.org.uk.
- ED Mynatt et al., “Aware Technologies for Aging in Place: Understanding User Needs and Attitudes,” Pervasive Computing (April–June 2004).
- A Sixsmith and N Johnson, “A Smart Sensor to Detect the Falls of the Elderly,” Pervasive Computing (April–June 2004).
- N Barnes et al., “Millennium Homes: A Technology Supported Environment for Frail and Elderly People,” in Proceedings of the Sixth Annual Scientific Conference of the Institute of Physics and Engineering in Medicine (York, UK: Institute of Physics and Engineering in Medicine, 2000).
This is an edited version of a paper originally presented at ICOST2006—4th International Conference on Smart Homes and Health Telematics, held in Belfast, Northern Ireland, UK, in June 2006.