Scalable AI Approach for Wi-Fi-Based Human Pose Estimation Using Compressed CSI Data

As the need for privacy-aware sensing grows, Wi-Fi-based human pose estimation is emerging as a viable alternative to vision-based systems. However, transmitting and processing large volumes of channel state information (CSI) poses a significant challenge, particularly for edge devices with limited resources. This project proposes a scalable AI-driven framework that compresses CSI data using vector quantization, significantly reducing bandwidth requirements without compromising pose estimation accuracy. To further optimize performance, a clustering-based method dynamically adjusts compression levels, and a Transformer model is integrated to enhance reconstruction under unstable network conditions. The system will be deployed on edge computing platforms such as Jetson Nano and Raspberry Pi, and evaluated in terms of latency, energy consumption, and pose estimation quality. The aim is to enable real-time, low-power, and privacy-preserving human sensing suitable for smart environments.