Wireless Sensing-Based Remote Detection of Concealed Metallic Objects

Abstract

Concealed weapon detection has gained widespread interest in recent times due to prevalent law and order situation. There is an ever-increasing need of detection of body-worn harmful objects from a safe distance to save precious human life and to cause minimal damage to infrastructure. Most of the conventional weapon detection schemes require proximity to target for detection. To alleviate the problem of close proximity, WiFi-based wireless sensing has recently emerged as a promising technique for remote detection and sensing in different applications. The focus of this research is to implement a low cost and robust WiFi-based wireless detection system for body-worn concealed objects. The methodology is based on utilizing low-cost WiFi sensors to acquire Channel State Information (CSI), smoothening/filtering of CSI data, extraction of different statistical parameters and building a model to differentiate among two cases of body-plus-weapon and body only. Probability density functions of the variance of CSI features are computed under metal and non-metal scenarios, that are non-overlapping, which validate the effectiveness of proposed approach to separate metal and non-metal scenarios. Furthermore, heatmap images are generated and a deep learning model is trained to automate the detection process. Our deep learning-based automated detection methodology has achieved an overall accuracy of 90.5% and 87.5% on test samples, respectively, for the detection of a metallic plate and a real gun in indoor setting at 20 ft distance from the antenna.