Seeing Through WiFi: Lightweight Human Pose Estimation with Dynamic Kernel Attention

2026-07-03

AI Summary

This work addresses the challenge of privacy-preserving, wearable-free human pose estimation in resource-constrained environments by proposing WiLHPE, a lightweight framework that leverages only WiFi channel state information (CSI) for efficient pose inference. During training, the model is guided by a visual modality, while at test time it operates solely on WiFi signals. The approach innovatively integrates dynamic learning convolutional kernels with a dual attention mechanism operating across both channel and frequency domains, substantially enhancing signal representation without increasing computational complexity. Hyperparameter optimization is efficiently handled via a Tree-Structured Parzen Estimator. Evaluated on the MM-Fi and WiPose datasets, the method achieves PCK50 accuracies of 85.96% and 94.27%, respectively, and maintains approximately 80% performance under Gaussian noise, demonstrating strong robustness and practicality.

Abstract

WiFi-based human pose estimation (HPE) enables the detection and interpretation of human body positions and movements without the need for wearable devices while preserving individual privacy concerns. Implementing this solution requires enhancing model performance and maintaining efficiency, especially on resource-constrained devices. This paper introduces a novel framework, WiLHPE, for lightweight and efficient human pose estimation using WiFi CSI signals. Empowered by a camera-based model during training, WiLHPE processes raw WiFi signals directly to estimate human poses in the testing phase. It employs a novel neural network architecture to dynamically learn convolutional kernels and apply attention mechanisms across channel and frequency spaces. This innovative method diversifies the kernels to improve the recognition capabilities of WiFi signals without adding complexity, ensuring efficiency. Additionally, the Tree-Structured Parzen Estimator algorithm is employed to optimize the critical hyperparameters of the neural network efficiently, minimizing the time required for optimal hyperparameter search compared to heuristic methods. Results from experiments on both the MM-Fi and WiPose datasets highlight the superiority of WiLHPE over state-of-the-art approaches, achieving 85.96% and 94.27% at PCK50, respectively, with minimal computational overhead. Notably, WiLHPE performs impressively even under challenging conditions, maintaining around 80% at PCK50 under AWGN noise with an error variance of 0.5.

Problem

• WiFi-based human pose estimation
• lightweight model
• privacy-preserving sensing
• resource-constrained devices
• CSI signals

Innovation

• Dynamic Kernel Attention
• Lightweight Pose Estimation
• WiFi CSI
• Channel-Frequency Attention
• Hyperparameter Optimization