CMU WiFi system maps bodies through walls, no camera

Using commodity WiFi hardware for through-wall human pose estimation is a legitimately impressive research result — and a double-edged privacy story hiding inside a "health monitoring" frame.

• –The architecture pipes raw CSI amplitude/phase through a CNN encoder-decoder to generate 2D feature maps, then feeds those into a modified DensePose-RCNN head — transfer learning bridges the domain gap between image and radio features
• –Performance is reported as comparable to camera-based DensePose, which is a strong claim given the fundamentally noisier input modality
• –The "privacy-preserving" framing (no camera footage) immediately inverts: any WiFi router in a home could theoretically become a covert through-wall tracker, a concern commentators noted widely after the paper's 2023 release
• –No official code release from CMU; multiple third-party reimplementations exist on GitHub including one adding real-time vital sign monitoring
• –Practical deployment still requires 3 transmitters and 3 aligned receivers, limiting real-world stealth scenarios for now