WiFireNet Architecture
A group of AI researchers at the Swedish Defense Research Agency has developed a system that can identify concealed handguns using wireless signals, offering a non-intrusive alternative to optical cameras and physical metal detectors. Detailed in a research paper, the system, called WiFireNet, analyzes channel state information derived from how objects distort ambient wireless signals. The complete setup relies on a transmitter, a receiver, and the core neural network.
When standard wireless signals encounter physical obstructions, they penetrate and reflect in distinct patterns dictated by the object's geometry, orientation, and material composition. This collected channel state information can be processed to classify objects with high precision. Previous security research has demonstrated that standard routers can map human presence with high accuracy using similar signal analysis.
Training and Dataset Creation
To develop the model, the team constructed a custom dataset by simultaneously capturing optical video footage and reflected wireless signals from individuals carrying various items, ranging from mobile phones to firearms. The deep learning model was trained using these synchronized video clips to associate visual representations of concealed weapons with specific wireless signatures. To differentiate firearms from similarly shaped everyday objects, the training regime included recordings of plastic training weapons, metal water bottles, real firearms, and mobile devices.
Performance and Limitations
Following training, the system was able to detect concealed weapons using only wireless channel data, though accuracy varied. In controlled evaluations tracking up to four moving individuals in an indoor environment, the system reached an average detection precision of 72.5%. However, when tasked with pinpointing the exact location of the weapon on a carrier's body, precision fell to 53.8%. The neural network also experiences performance degradation when deployed in unfamiliar indoor layouts.
Future Development and Privacy
The prototype remains experimental and was constructed using a modified implementation of the older IEEE 802.11a wireless standard to establish a proof of concept. Future iterations plan to incorporate modern Wi-Fi 7 and Wi-Fi 8 networks, which provide higher-resolution channel data to improve detection reliability. The researchers have made the project open source, releasing the underlying code, trained models, and dataset. While wireless sensing is emerging as a surveillance technique that avoids complex optical hardware, critics note that the widespread availability of wireless infrastructure combined with minimal regulation creates significant privacy challenges.
