The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)2027

RadPose: ISAC-based RF 3D Human Pose Estimation

Geo Jie Zhou, KunZhe Song, Ruxin Lin, Huacheng Zeng

INSS Lab, Michigan State University

Demo Video

We demonstrate a video of ISAC data and the corresponding human pose estimation results, including typical movements in our dataset (walking, waving, squatting and standing).

Demo Video

Abstract

Human pose estimation under adverse lighting and occlusion conditions remains challenging for camera- and LiDAR-based systems. RadPose is a learning-based monostatic integrated sensing and communication (ISAC) framework that enables commodity mmWave communication devices to perform 3D human pose estimation. It extracts radio point clouds directly from standard OFDM communication signals and combines cross-frame pose inference with inverse-kinematics refinement for robust estimation in seen, unseen, and occluded environments.

Qualitative results

The following examples show RadPose predictions across seen and unseen environments.

Seen environment

RadPose qualitative results in a seen environment

Unseen environments

RadPose qualitative results in an unseen environment

Additional RadPose result in an unseen environment

Additional RadPose result in a second unseen environment

Additional RadPose result in a third unseen environment

Citation

@inproceedings{zhou2027radpose,
  title={RadPose: ISAC-based RF 3D Human Pose Estimation},
  author={Zhou, Geo Jie and Song, Kunzhe and Lin, Ruxin and Zeng, Huacheng},
  booktitle={The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2027}
}