The research team at NYU Tandon has developed an ensemble approach that combines multiple cutting-edge AI algorithms. Rather than relying on a single AI model that could mistake a red car or a sunset for a fire, the system requires matching between multiple algorithms before checking fire detection, significantly reducing false alarms, a critical consideration in emergencies.
Researchers trained the models by constructing a comprehensive custom image data set representing all five classes of fires recognized by the National Fire Protection Association, ranging from regular flammable materials to electrical fires and cooking-related incidents. The system achieved a significant accuracy rate, with the best performance model combination reaching detection accuracy of 80.6%.
The system incorporates temporal analysis to distinguish between actual fires and static fire-like objects that can cause false alarms. By monitoring how the size and shape of detected fire areas change in successive video frames, the algorithm can distinguish between static images of actual growing fires and flames hanging from the wall. “The real fire is dynamic, growing and transformed form,” explained Sunil Kumar, May's professor. “Our system tracks these changes over time and achieves 92.6% accuracy in eliminating false positives.”
The technology runs within a cloud-based, thing-based architecture where multiple standard security cameras stream RAW video to a server that performs AI analysis. When a fire is detected, the system automatically generates video clips and sends real-time alerts via email and text messages. This design means that the technology can be implemented using existing CCTV infrastructure without the need for expensive hardware upgrades. This is an important advantage for widespread adoption.
This technology can be integrated into drones or unmanned aerial vehicles to search for wildfires in remote forest areas. Early-stage wildfire detection prioritizes evacuation orders that purchase important times in the race to contain and destroy them, allowing faster deployment of resources and dramatically reduce ecological and property losses.
The same detection system can be embedded in tools that already carry the firefighters to improve safety and assist in fire response: helmet cameras, thermal imagers, vehicle mounted cameras, and autonomous fire robots. In urban areas, UAVs integrated with this technology can help you perform a 360 degree size up, especially if there is a fire on higher floors of high-rise structures.
“It can be remotely assisted in identifying the location of the fire and the potential for trapped residents,” said Colonel John Celiero of the New York City Fire Department.
Beyond fire detection, researchers point out that it could adapt approaches to other emergency scenarios, such as security threats and medical emergencies, and could expand the way society monitors and responds to a variety of safety risks.
In addition to Panindre and Kumar, the research team includes Nanda Kalidindi ('18 MS Computer Science, NYU Tandon), Shantanu Acharya ('23 MS Computer Science, NYU), and Praneeth Thummalapalli ('25 MS Computer Science, NYU Tandon).
