RF-Care: RFID-Based Human Pose Estimation for Nursing-Care Applications Under Visual Occlusion
Jixiao Liu, Zihou Xia, Junyi Liu, Shijie Guo · IEEE Sensors Journal · 2024
Occlusion and environmental light changes often exist in intelligent care scenarios. Human posture sensing through vision alone is prone to misjudgment, so human posture sensing methods other than vision are needed. This article proposes RF-Care, an RFID-based human pose estimation system that avoids these problems. RF-Care uses the phase difference between RFID tags on the limbs and antennas to determine body joint positions and motion. It requires only a wearable RFID tag array and a simple RF antenna setup. We introduce a fine-grained limb tracking algorithm and dual-antenna 3-D pose estimation method to achieve accurate results despite occlusions. Experiments validate RF-Care’s performance for subjects standing or lying in bed under blankets/quilts. Compared to Kinect 2.0, the mean position error is about 4 cm without occlusion. The standard deviation (STD) is about 3.4 cm when visually occluded. RF-Care efficiently monitors full-body posture regardless of impediments, benefiting intelligent nursing care and other applications requiring accurate real-time pose measurement.