WiFireNet Detects Concealed Handguns Using Wi-Fi Signals
TechNews reports that the Swedish Defence Research Agency (FOI) developed WiFireNet, an experimental AI system that identifies concealed handguns from changes in Wi-Fi channel state information (CSI). In controlled tests in a small room, the system tracked up to four moving people and achieved 72.5% average accuracy for handgun detection, while accuracy for locating a weapon on a carrier fell to 53.8%. The arXiv preprint also reports unstable performance in unfamiliar room layouts and describes the implementation as a proof of concept using modified IEEE 802.11a equipment; the code, model, and WiFire dataset were open-sourced for further research.
WiFireNet demonstrates an open-source, camera-free approach to concealed-weapon detection, but its reported accuracy depends on controlled conditions and weakens when the task or environment changes.
瑞典國防研究局（FOI）開發出名為 WiFireNet 的實驗性人工智慧系統，嘗試用 Wi-Fi 訊號變化辨識隱藏的手槍。論文 13 日以預印本發表於 arXiv，提供另一種不必依賴攝影機或金屬探測門的偵測方法。
WiFireNet 核心技術在分析所謂「信道狀態資訊」（Channel State Information，CSI），即物體如何改變 Wi-Fi 訊號的測量結果。團隊指出，當 Wi-Fi 訊號遇到物體，會因形狀、方向與材質不同產生不同穿透與反射特徵，或可辨識物體。為了訓練模型，研究員建立 WiFire 資料庫，同步錄製 RGB 影像與反射的 Wi-Fi 訊號，並讓受試者攜帶手機、塑膠手槍、金屬水瓶與真實金屬手槍並在室內移動，以協助模型學會區分外形或材質相近的物品。
#Swedish Defence Research Agency (FOI)#WiFireNet#WiFire#IEEE 802.11a#Wi-Fi 7#Wi-Fi 8