3D Human Pose Estimation for Free-from and Moving Activities Using WiFi (2204.07878v1)

Abstract: This paper presents GoPose, a 3D skeleton-based human pose estimation system that uses WiFi devices at home. Our system leverages the WiFi signals reflected off the human body for 3D pose estimation. In contrast to prior systems that need specialized hardware or dedicated sensors, our system does not require a user to wear or carry any sensors and can reuse the WiFi devices that already exist in a home environment for mass adoption. To realize such a system, we leverage the 2D AoA spectrum of the signals reflected from the human body and the deep learning techniques. In particular, the 2D AoA spectrum is proposed to locate different parts of the human body as well as to enable environment-independent pose estimation. Deep learning is incorporated to model the complex relationship between the 2D AoA spectrums and the 3D skeletons of the human body for pose tracking. Our evaluation results show GoPose achieves around 4.7cm of accuracy under various scenarios including tracking unseen activities and under NLoS scenarios.

Overview of "3D Human Pose Estimation for Free-form and Moving Activities Using WiFi"

The paper "3D Human Pose Estimation for Free-form and Moving Activities Using WiFi" introduces GoPose, a 3D pose estimation system leveraging conventional WiFi devices. Unlike traditional approaches that utilize specialized equipment such as cameras or wearables, GoPose capitalizes on the ubiquitous presence of WiFi in home settings to facilitate mass adoption without additional costs.

The WiFi-based system utilizes the two-dimensional Angle of Arrival (2D AoA) spectrums derived from channel state information (CSI) to ascertain spatial details about human body parts, permitting pose estimation independent of environmental constraints. Coupling these spectrums with deep learning techniques—specifically convolutional neural networks (CNNs) and Long Short-Term Memory (LSTM) networks—the system maps the spectrums to construct 3D skeleton models of human poses.

Methodology and Contributions

The approach is premised on harnessing WiFi signals reflected off human bodies to track dynamic 3D poses. The system addresses multiple challenges:

1. Spatial Information Extraction: Without direct spatial data, the 2D AoA is derived using the spatial diversity of antennas and the frequency diversity of OFDM subcarriers, enhancing resolution and aiding in localization of various body parts.
2. Environment Independence: GoPose dismisses static environmental impacts by subtracting these constant interferences from the dynamic spectrum data during activity periods.
3. Modeling Complexity: Employing CNNs to process the spatial layout of body parts and LSTM networks to model temporal dynamics of human motion, the intricate mappings of the 2D AoA spectrums to 3D skeletons are effectively constructed.

The empirical evaluations demonstrate that GoPose achieves precision with an average error margin of 4.7 cm, applicable under line-of-sight (LoS) and non-line-of-sight (NLoS) conditions, and performs reliably for diverse and unforeseen activity scenarios.

Results and Implications

GoPose showcases significant advantages in scenarios where traditional vision-based methods falter, such as low-light conditions and occlusions. The practical implications for smart homes are multifaceted, offering potential enhancements in domains such as virtual/augmented reality, interactive gaming, and health monitoring, all without privacy concerns typically associated with camera systems.

Theoretically, this work exemplifies how commodity technology like WiFi can be innovatively repurposed to extend beyond its conventional utility, presenting an accessible and cost-effective solution for 3D human pose estimation.

Future Directions

Looking forward, the paper indicates potential expansion opportunities, particularly in increasing system scalability for environments with more than two users. Expanding the dataset with a larger pool of subjects, especially in diverse setups, would likely increase the robustness and reliability of the GoPose model. Additionally, further enhancements in the resolution of AoA spectrums via hardware advancements could proportionately improve long-term detection range and precision.

The integration of GoPose in smart home environments underscores the transformative role of ambient technologies in pervasive computing and human-computer interaction paradigms. Continued research in this vein is poised to further bridge the gap between passive sensing capabilities and active interaction models without impeding user privacy or convenience.