Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph-based 3D Human Pose Estimation using WiFi Signals

Jichao Chen, YangYang Qu, Ruibo Tang, Dirk Slock

TL;DR

The paper tackles 3D human pose estimation from WiFi CSI by explicitly modeling skeletal topology. It introduces GraphPose-Fi, a pipeline that combines a shared CNN encoder across antennas, a lightweight temporal-spatial attention module, and a graph-based regression head using Chebyshev graph convolutions with self-attention to capture local bone connections and global joint dependencies. On the MM-Fi Protocol 1 dataset, GraphPose-Fi achieves state-of-the-art performance across random, cross-subject, and cross-environment splits, with clear improvements in MPJPE and PCK, particularly for torso and head joints, while hands/elbows remain challenging due to limited spatial resolution. Ablation studies validate the LTSA design and the graph-based regression head, and the work suggests future directions in multi-sensor fusion and robust domain generalization.

Abstract

WiFi-based human pose estimation (HPE) has attracted increasing attention due to its resilience to occlusion and privacy-preserving compared to camera-based methods. However, existing WiFi-based HPE approaches often employ regression networks that directly map WiFi channel state information (CSI) to 3D joint coordinates, ignoring the inherent topological relationships among human joints. In this paper, we present GraphPose-Fi, a graph-based framework that explicitly models skeletal topology for WiFi-based 3D HPE. Our framework comprises a CNN encoder shared across antennas for subcarrier-time feature extraction, a lightweight attention module that adaptively reweights features over time and across antennas, and a graph-based regression head that combines GCN layers with self-attention to capture local topology and global dependencies. Our proposed method significantly outperforms existing methods on the MM-Fi dataset in various settings. The source code is available at: https://github.com/Cirrick/GraphPose-Fi.

Graph-based 3D Human Pose Estimation using WiFi Signals

TL;DR

The paper tackles 3D human pose estimation from WiFi CSI by explicitly modeling skeletal topology. It introduces GraphPose-Fi, a pipeline that combines a shared CNN encoder across antennas, a lightweight temporal-spatial attention module, and a graph-based regression head using Chebyshev graph convolutions with self-attention to capture local bone connections and global joint dependencies. On the MM-Fi Protocol 1 dataset, GraphPose-Fi achieves state-of-the-art performance across random, cross-subject, and cross-environment splits, with clear improvements in MPJPE and PCK, particularly for torso and head joints, while hands/elbows remain challenging due to limited spatial resolution. Ablation studies validate the LTSA design and the graph-based regression head, and the work suggests future directions in multi-sensor fusion and robust domain generalization.

Abstract

WiFi-based human pose estimation (HPE) has attracted increasing attention due to its resilience to occlusion and privacy-preserving compared to camera-based methods. However, existing WiFi-based HPE approaches often employ regression networks that directly map WiFi channel state information (CSI) to 3D joint coordinates, ignoring the inherent topological relationships among human joints. In this paper, we present GraphPose-Fi, a graph-based framework that explicitly models skeletal topology for WiFi-based 3D HPE. Our framework comprises a CNN encoder shared across antennas for subcarrier-time feature extraction, a lightweight attention module that adaptively reweights features over time and across antennas, and a graph-based regression head that combines GCN layers with self-attention to capture local topology and global dependencies. Our proposed method significantly outperforms existing methods on the MM-Fi dataset in various settings. The source code is available at: https://github.com/Cirrick/GraphPose-Fi.

Paper Structure

This paper contains 13 sections, 7 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Preliminary
  4. Proposed Method: GraphPose-Fi
  5. Per-Antenna Feature Encoder
  6. Lightweight Temporal and Spatial Attention
  7. Graph-based Pose Regression
  8. Training Objective
  9. Experiments
  10. Experimental Setup
  11. Comparison with State-of-the-art Methods
  12. Ablation Study
  13. Conclusion

Figures (1)

  • Figure 1: The overview of our proposed GraphPose-Fi for WiFi-based 3D HPE. It consists of a per-antenna feature encoder, a lightweight temporal and spatial attention module, and a graph-based pose regression head.