3D Human Pose Tracking with RFID

RF-based human pose estimation has attracted considerable interest recently as an effective means of human-computer interaction (HCI). Compared with camera-based alternatives, RF sensing can better protect user’s privacy and are robust to lighting, cluttered background, view angle, and non-line-of-sight conditions. However, due to complicated indoor propagation environments, most RF-based sensing methods are sensitive to the deployment environment, are hard to generalize, and rely heavily on large amounts of training data. In this talk, we present RFID-Pose, which uses RFID tags as wearable sensors to detect 3D human pose. In addition to introducing the RFID-Pose system, we also investigate various enhancements for (i) making the model more generalizable to various subjects and environments, (ii) making the model more generalizable to different wireless technologies, and (iii) reducing the dependence on large amounts of training data via data augmentation. Experiments conducted in various environments demonstrate the high performance and efficacy of the proposed approaches.