Swedish scientists at the Defence Research Agency (FOI) have demonstrated that everyday WiFi signals can help identify people carrying concealed handguns inside a room. Their experimental system analyzes how radio waves bounce off and pass through objects and bodies, offering a potential complement to cameras and metal detectors.
Key Takeaways:
- Swedish Researchers develop modern technology that uses WiFi to detect hidden guns and alert security
- WiFi setup can distinguish who carries a hidden gun, and who has a water bottle
- WiFi sensing addresses some of these gaps. Signals already fill most indoor environments, can penetrate clothing, and do not require people to queue up
The work, detailed in a preprint titled “Concealed Handgun Detection using WiFi Signals” and released in mid-October 2026, introduces both a new dataset called WiFire and a specialized deep-learning model named WiFireNet.
Why Current Security Tools Fall Short
Traditional camera systems struggle with concealment, poor lighting, or physical obstacles. Metal detectors require people to pass single-file through narrow gates, creating bottlenecks and limiting their use in open or crowded spaces. They also cannot easily distinguish a handgun from other metal objects such as phones or bottles.
WiFi signals can help detect concealed handguns. Image by Cybernews.
WiFi sensing addresses some of these gaps. Signals already fill most indoor environments, can penetrate clothing, and do not require people to queue up. The approach relies on Channel State Information (CSI)—detailed measurements of how WiFi waves change in amplitude and phase as they travel through a space. Different materials and shapes distort the signal in distinctive ways.
How the Experiment Worked
Researchers set up two WiFi transmitters on one side of a room and two receivers on the other. They used a modified open-source implementation of the older IEEE 802.11a standard operating at 5.68 GHz with a 20 MHz bandwidth—frequencies common in many existing routers. CSI data was collected at 80 samples per second while a standard RGB camera recorded synchronized video for labeling.
The setup used for data collection. Screenshot from the “Concealed Handgun Detection using WiFi Signals” study.
Tests took place in two indoor spaces: an 8 m² room with 2 people and a larger 12 m² room (up to 4 people moving freely). Participants carried common everyday items including mobile phones, metal water bottles and, in some cases, a deactivated metal handgun or a plastic replica of similar shape. The system had to detect and localize the objects while distinguishing real threats from look-alikes.
The setup used for data collection. Screenshot from the “Concealed Handgun Detection using WiFi Signals” – Photo Credit: Cyber News
The team created the WiFire dataset containing aligned WiFi CSI samples and RGB annotations. They then trained WiFireNet, a deep-learning model that first translates the one-dimensional CSI phase and amplitude data into a two-dimensional image-like representation using a novel convolutional Modality Translation Network. From there, the model performs object detection similar to visual systems, producing bounding boxes and class predictions.
Performance Results
In baseline tests the system achieved object recall (the percentage of actual carried items it found) between roughly 83% and 91.5%, depending on room size and the number of people present. Mean average precision scores, which penalize false positives and imprecise localization, ranged from about 35% to 50%.
For the handgun class specifically, average precision varied more widely—reaching as high as 72.5% under a lenient localization threshold in the larger room with the standard WiFi configuration, and dropping to around 53.8% under stricter criteria. The model performed best with the conventional IEEE 802.11a setup in some cases and showed that it could operate in multi-person dynamic scenes, something earlier WiFi object-detection studies rarely attempted.
The researchers emphasize that these figures demonstrate feasibility rather than a ready-to-deploy product. The system can indicate that a person is carrying a suspicious object but does not yet pinpoint the exact location of the weapon on the body with high precision.
Limitations and Open Questions
The experiments used controlled laboratory rooms and modified research-grade radio hardware rather than ordinary consumer routers. Performance in unfamiliar environments, larger crowds, or real-world multipath conditions remains untested. Privacy concerns also arise: a passive WiFi-based detector could theoretically screen people without their knowledge.
The team has released the full WiFire dataset, model weights, and code on Hugging Face and GitHub so other laboratories can reproduce, critique, or improve the work.
Future Potential
WiFi sensing has already been explored for activity recognition, pose estimation, and presence detection. Extending it to concealed-weapon identification in multi-person settings is new. The FOI researchers position the technology as a complementary layer that will be able to cover areas where cameras are blocked or metal detectors are impractical rather than a full replacement for existing tools.
Further advances in CSI resolution, multi-antenna setups, and more diverse training data will be needed before such systems could be considered for practical security applications. For now, the Swedish study shows that the radio waves already filling our buildings carry more information than previously assumed—and that, with the right algorithms, they can help spot what the eye cannot see.
