Presentation Information
[B-19-09]A Preliminary Study on Non-Contact Heart Rate Estimation Using Wi-Fi CSI with Deep Fusion and ResNet-50
◎△Shoei Sato1, Takeshi Toda2 (1. Graduate School of Science & Technology, NIHON UNIVERSITY, 2. College of Science & Technology, NIHON UNIVERSITY)
Keywords:
HAR,Wi-Fi,CSI,Healthcare,Deep Learning
In Japan’s aging society, ensuring the safety of elderly people living alone and residents of care facilities is a critical issue. To address this challenge, low-cost, privacy-conscious, contactless vital sign estimation using Wi-Fi networks has garnered attention. By utilizing the Channel State Information (CSI) feedback function of IEEE 802.11ac, it is possible to estimate biometric information—such as a person’s heart rate—non-invasively based on the radio wave propagation characteristics of a given space. In our previous work, we proposed a method that dynamically combines CSI amplitude and phase features using the Deep Fusion technique to estimate BPM (heart rate) via ResNet-50. In this paper, we report on our evaluation of how differences in subject location across eight indoor locations affect estimation accuracy.