Low cost fire detection uses AI on WiFi signals – video
Engineers in Australia have tested AI on Raspberry Pi single board computers in a low cost fire detection system that monitors the changes in WiFi signals.
A controlled test detonation of a car by the Sydney Harbour Tunnel Company provided further data to demonstrate the effectiveness of the technology.
Professor Aruna Seneviratne, Dr Deepak Mishra, and a team from the School of Electrical Engineering and Telecommunications at the University of New South Wales (UNSW) identified the distinctive patterns in data from radio signals during fire events using artificial intelligence to analyse the environment in real-time. This picks up on changes to the WiFi signals as they pass through the higher temperature air and smoke.
The system can then determine with greater accuracy whether any atmospheric changes are being caused by a real fire, and if so, raise an alarm or trigger an automatic sprinkler system.
- WiFi fights back in the Internet of Things
- Coating boosts WiFi reception through walls
- First live demo of WiFi 7 technology
Existing detection systems, which are largely based on thermal imaging, often produce false positive readings by detecting levels of smoke or changes in temperature which are not dangerous or caused by an actual fire – perhaps from a faulty exhaust pipe on a vehicle, a hot radiator or even flickering lights.
But Seneviratne and his team were able to showcase the technology during a controlled test in the middle of the night inside the Sydney Harbour Tunnel.
“Existing specialised fire detection cameras can cost around $10,000 to buy, whereas our transmitters and receivers are $100 or even less,” said Seneviratne. “The other thing with cameras is that they need to be carefully maintained. The lenses need to be kept clean and they often need to be properly aligned. With our system, the transmitters and receivers are just sending out a radio signal and there is very little maintenance required. Therefore, there is also a much lower cost to operating the system.”
The test with Trantek MST, the systems vendor for the tunnel, and the tunnel owner/operator used a series of Raspberry-Pi-based WiFi transmitters and receivers to monitor the environment as a test car prepared for the purpose was detonated and set on fire during a scheduled emergency response training exercise.
- Machine learning tool ported to Raspberry Pi 4
- Kontron looks to Raspberry Pi4 move for industrial AI
This uses the variation in the different subcarrier frequencies of WiFi, amalgamating the effect of the environmental phenomena on all the subcarrier frequencies and finding the most sensitive frequencies which help aid the analysis. Up to 1300 packets of data per second can be processed and analysed.
“As the air temperature changes, so does its density, and that changes the signature of the reading when we receive the signal. In fact, we have experimentally demonstrated that these changes are strongly correlated with the rise or fall of temperature in the environment between transmitter and receiver,” said Seneviratne.
“Smoke and different gases, such as carbon monoxide that can be produced in fire situations, also affect the density of the air and will give distinctive signatures on our readings. Specifically, these signatures are captured in the form of wireless channel information,” he said, “What we also add into the system is artificial intelligence to analyse all the data and compare to baseline readings to help determine if there is a real fire occurring.”
The new system developed at UNSW exploits the fact that Wi-Fi waves have various transmission frequencies, known as subcarriers. Just as different wavelengths of light are affected uniquely by different objects
The technique is important to improve confidence in automatic fire detection systems which can currently sometimes struggle even to differentiate between a fire and a bright flickering neon light. The UNSW team believe their system can be applied in a wide range of environments, including in industrial locations, commercial high-rise buildings and even in the home.
Assembling an array of transmitters and receivers also helps to identify the zonal location of a specific fire which can then aid emergency services to respond quickly and efficiently.
“Traditional sensor methods aren’t effective or regularly detect false positives and the facility management operators can’t discriminate when a real fire emergency is in progress,” said Leo Ascone, CEO of Trantek MST. “As a maker of high availability, distributed operations management and control systems Trantek MST is now set to combine the UNSW Wi-Fi breakthrough with video analytic technology and create a new era in fire safety operations, opening the door for diverse application deployments spanning transport, defence and industrial sites, as well as commercial and domestic buildings. The industry-led research collaboration between Trantek MST and UNSW has significantly accelerated the readiness for deployment. It’s now time to rewrite the standards on fire detection.”
Other board articles
- Okdo in exclusive deal to supply Rockchip boards
- Raspberry Pi warns of bots as supply still tight
- Competition to find the longest serving Raspberry Pi
Other articles on eeNews Europe
- Intel prepares for trillion transistor era shake up
- 1400 RISC-V cores for on-chip machine learning
- PAM4 chips for 800G active cables
- First RISC-V processor starts operation in orbit