Graph-based 3D Human Pose Estimation using WiFi Signals

2025-11-24

AI Summary

Existing WiFi-based 3D human pose estimation methods predominantly adopt end-to-end regression, neglecting anatomical joint topology constraints—leading to degraded performance under occlusion. To address this, we propose the first WiFi pose estimation framework that explicitly models the human skeletal graph structure. Our method employs a CNN encoder to extract multi-antenna time-frequency features, incorporates a lightweight cross-antenna and temporal self-attention mechanism to enhance discriminative signal representation, and integrates graph convolutional networks (GCNs) with channel-temporal feature reweighting for topology-aware end-to-end learning. Evaluated on the MM-Fi dataset, our approach achieves significant improvements over state-of-the-art methods, demonstrating exceptional robustness under occlusion and in complex environments. Moreover, it preserves user privacy and maintains computational efficiency—making it suitable for real-world deployment.

Technology Category

• Computer Vision: Biometrics, Face, Gesture & Pose
• Humans and AI: Interaction Techniques and Devices
• Application Domains: Humanities & Computational Social Science

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

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.

Problem

• Modeling skeletal topology for WiFi-based 3D human pose estimation
• Capturing local and global dependencies in human joint relationships
• Overcoming limitations of direct regression from WiFi signals to coordinates

Innovation

• Graph-based framework modeling skeletal topology
• CNN encoder with attention for feature extraction
• GCN and self-attention capturing joint dependencies