부산대학교 도서관

In this article, we study the feasibility of a novel idea by coupling radio frequency (RF) sensing technology with correlated knowledge distillation (CKD) theory toward designing lightweight, near real-time, and precise human pose monitoring systems through an in-house experimental testbed development at Deakin University. The datasets collected from the developed testbed are fed to the CKD framework, which transfers and fuses pose knowledge from a robust “Teacher” model to a parameterized “Student” model and can be a promising technique for obtaining accurate yet lightweight pose estimates. As a result, it becomes possible to identify the current pose of a human body (e.g., an elderly individual in a care home) solely using RF signals (e.g., WiFi signals in a home) without relying on visual signals or video information, meaning that this approach effectively addresses privacy concerns by allowing pose identification without compromising personal privacy. To ensure its efficacy, we implement CKD for distilling logits in our integrated software-defined radio (SDR)-based experimental setup and investigate the RF-visual signal correlation. Our CKD-RF sensing technique is characterized by two modes—a camera-fed teacher class network (e.g., images and videos) with an SDR-fed student class network (e.g., RF signals). Specifically, our CKD model trains a dual multibranch teacher–student network by distilling and fusing knowledge bases. The resulting CKD models are then subsequently used to identify the multimodal correlation and teach the student branch in reverse. Instead of simply aggregating their learnings, CKD training comprises multiple parallel transformations with the two domains, i.e., visual images and RF signals. Once trained, our CKD model efficiently preserves privacy and utilizes the multimodal correlated logits from the two different neural networks (NNs) to estimate poses solely using RF signals.