RF-Care: RFID-based Human Pose Estimation for Nursing-care Applications

Abstract

Continuously and accurately measuring the human posture of the care receiver while the human body is partially obscured, is one of the major challenges in many future smart nursing-care scenarios. To address human posture estimation when visual information is not available, we introduce an RFID-based method to estimate human pose without introducing extra burden to the subject, named the "RF-Care" system. Multiple wearable RFID tags on human clothing and a dual-antenna reader system are used to extract multiple phase features to estimate the 3D pose of the human body. We developed an NLOS (Non-Line of Sight) human posture estimation method to estimate limb angles by modeling the signals of multiple wearable tags moving in space and then combined with a human skeleton model to reconstruct the human posture. Experimental data validate the developed method and system, and show that the RFID system can effectively monitor human posture in real time, especially when the human body is covered by a blanket. When uncovered, the average angular error of the limbs is about 6°, or the average positional error of the limbs is about 4 cm with Kinect 2.0 as the standard; when covered, the angular standard error of the limbs is about 6°, or the positional standard error of the limbs is about 4 cm. The developed system would benefit various nursing-care applications such as nursing-care robots and untethered monitoring at home.