Fine-Grained Human Imaging Based On Commercial Wi-Fi Signals

Wireless sensing technology captures wireless radio frequency signals in the environment to perceive human motion and behavior,enabling a variety of requirements to be met without the need for additional sensors to be deployed.Currently,wireless sensing has achieved many results in the field of human activity recognition,but existing solutions only provide coarse-grained information.The commonly used cameras also have problems such as privacy leakage,line of sight obstruction,and the need for light,which often limit their application.Wireless imaging can effectively solve these problems.Therefore,this article focuses on using wireless signals for fine-grained human body imaging tasks and provides an effective alternative in scenarios where some cameras cannot work or work is restricted.This article explores and researches fine-grained human body imaging based on commercial WiFi signals from the perspective of practical applications.A fine-grained human body imaging system based on commercial WiFi signals is designed,which uses the Atheros csi toolkit and Qualcomm hardware platform to collect channel state information(CSI)and timestamp information under 5GHz.Compared with received signal strength,CSI is a complex signal with multiple carriers and channels,providing more environmental information.By using effective CSI signals to train deep learning models,the mapping from the signal domain matrix to the video is achieved,completing the fine-grained human body imaging video task in multi-angle and multi-objective scenes.The feasibility and superiority of the solution are confirmed through extensive on-site data measurement and evaluation,achieving the following research results:This article proposes a novel multi-angle solution based on wireless channel state information,which solves the problem of WiFi fine-grained human body imaging in multiangle and multi-objective scenes,and completes the dual-view wireless imaging task in different scenes by training a deep learning model.This article combines commercial WiFi equipment to build a signal acquisition platform and proposes an algorithm for constructing a multi-source heterogeneous dataset using wireless channel state information and video.To solve the problem of random loss of radio frequency signals in the wireless signal perception field,an anomaly exclusion and preprocessing algorithm is designed to enhance data quality by processing the collected signal and video data.After preprocessing,a time matching algorithm is designed to integrate multi-source heterogeneous data to serve wireless imaging tasks.A dual-view fine-grained human body imaging algorithm based on deep learning is proposed.Due to the difficulty in mapping from the signal to the video,traditional signal analysis methods cannot explore the mapping relationship.Therefore,this article designs three algorithm modules to extract signal features,map signal features to image features,and restore dual-view video frames to complete the human body imaging task in dual views.Finally,performance evaluation of the network model was conducted using a multisource heterogeneous test set constructed with wireless channel state information,confirming the feasibility and robustness of wireless signal-based video generation technology with a maximum average intersection over union of 0.83.This technology has a promising market outlook and can provide support for further research in wireless sensing technology.