Researchers have developed WiFireNet, an experimental AI system capable of detecting concealed handguns by analyzing Wi-Fi channel state information. The model was trained on the WiFire Dataset, which includes synchronized RGB video footage and Wi-Fi signal reflections of individuals carrying various objects, including plastic and metal handguns, phones, and water bottles.
In controlled indoor tests involving up to four moving subjects, the system reached an average detection precision of 72.5%. When tasked with pinpointing the exact location of a weapon on a person, the precision decreased to 53.8%. The researchers noted that the model currently faces performance challenges when operating in unfamiliar indoor environments.
The prototype was built using the IEEE 802.11a wireless standard from 1999. Future development plans involve transitioning to Wi-Fi 7 and Wi-Fi 8 networks to leverage higher-resolution channel state information. The project has been released as open source, providing access to the code, models, and dataset for further research.
The technology utilizes the way objects distort wireless signals based on their material, shape, and orientation. While Wi-Fi sensing is emerging as a surveillance method that avoids the need for specialized hardware, the researchers acknowledged that the widespread availability of such technology raises significant privacy concerns.
