Researchers from the Swedish Defence Research Agency (FOI) have introduced WiFireNet, an AI system designed to detect concealed handguns using Wi-Fi signals. The study, published on arXiv on October 13, details how the system analyzes Channel State Information (CSI) to identify objects based on how they disrupt wireless signals.
The system architecture consists of a Wi-Fi transmitter, a receiver, and the WiFireNet AI model. To train the model, the team created the WiFire Dataset, which includes synchronized RGB video and Wi-Fi signal data for various items, such as mobile phones, metal cups, plastic guns, and real metal handguns.
In controlled indoor tests tracking up to four people, the system achieved an average detection accuracy of 72.5%. However, accuracy dropped to 53.8% when attempting to pinpoint the exact location of the weapon on a person. Performance also decreases significantly in unfamiliar indoor environments.
The current prototype utilizes the outdated IEEE 802.11a wireless standard. Future development plans include upgrading to Wi-Fi 7 and Wi-Fi 8 to obtain higher-resolution CSI data. The research team has made the code, model, and dataset open-source to facilitate further study in the field.
