NYU’s Fire Research Group at Tandon School of Engineering developed an artificial intelligence system that enables security cameras to detect fires within 0.016 seconds per frame.
The team’s study, published in the IEEE Internet of Things Journal, began when a firefighter from the Fire Department of New York raised concerns about deaths due to undetected fires. Awareness of the issue was spurred by a 2024 report from the National Fire Protection Association, which found that 28% of residential fire fatalities occurred when alarms failed to alert individuals.
Prabodh Panindre, the study’s lead researcher and an assistant research professor at Tandon, told WSN that the visual fire-detection system helps reduce the blind spots in traditional smoke and carbon dioxide detectors by immediately processing surveillance feeds on CCTV cameras through multiple AI algorithms.
“We have several public spaces like parks and forests where we don’t have any smoke alarms or anything,” Panindre said. “We can make use of the CCTVs, or all these camera recording devices everywhere throughout our infrastructure, by detecting security threats like fire and smoke.
In the study, researchers tested the AI model using the FDNY report to simulate five classes of fires defined by the NFPA. Once certain characteristics — such as smoke and fire patterns — are detected, it immediately notifies the server. The model’s best performing iteration reached an accuracy rate of 80.6%.
“We have been trying to increase our data set and improve our accuracy, making sure that we can address and detect all types of fires,” Panindre said.
Because the detection process previously relied on human input, the AI system was able to significantly expedite responses. Panindre said that a new approach had long been needed for security systems, and that improper maintenance and installment of traditional detectors are often why fires get out of control.
Compared to traditional fire detective systems, CCTV can also cover a larger field of view, allowing firefighters to respond to fires in places that lack chemical detection systems. The existing network of privately-owned CCTV cameras allows for areas both indoors and outdoors to have rapid fire detection capabilities.
This development is a part of NYU Fire Research’s broader efforts to improve firefighting technology through collaborations with fire services across the United States. Panindre stressed the importance of the group’s relationship with the FDNY in spreading crucial techniques and research across the fire departments throughout the country.
“Our focus is not just doing the research, but on translation of research,” Panindre said. “We need to make sure our end user, the firefighter, is able to understand the research and is able to implement that in their real life practices.”
Contact Alex Amaral at [email protected].