WiSECam: A CSI-Based Deep Learning Motion Detection for Wireless Cameras

초록

Nowadays surveillance cameras play an important role in safety and security. In particular, a Wi-Fi camera is gaining popularity because of its easy installation. However, there is a visibility limitation of existing commodity cameras, for instance, they can not see an object behind an occlusion. In addition, environmental factors, such as dust, lighting, and humidity can also adversely affect the camera's visibility. To address these problems, the prevalent solutions use multiple cameras to cover the entire scene. But it comes with a cost. We, therefore, propose a cheaper, responsive, and effective solution called WiSECam (Wi-Fi-Security Enhanced Camera), which can be used as stand-alone in such as bank lockers, museums, and goldsmith stores that are proneto burglary. Our proposed solution leverages the Channel State Information (CSI) from Wi-Fi signals to detect human motion. We devise a deep learning model by using the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to handle the sequential CSI data. We implemented and evaluated it in the considered real-life scenarios. WiSECam achieves an average accuracy of around 98% with 1 second response time in different real-life settings. © 2021, Korean Institute of Communications and Information Sciences. All rights reserved.

키워드