davidakpele/wifi-densepose

What it solves

WiFi DensePose provides a way to perform human pose estimation without using cameras. By leveraging WiFi signals, it offers a privacy-preserving alternative for tracking human movement and posture in real-time, which is particularly useful for environments where cameras are intrusive or prohibited.

How it works

The system uses Channel State Information (CSI) from standard WiFi routers and access points. It processes these signals through a pipeline consisting of a CSI Data Collector, a Phase Sanitizer to remove noise, and a DensePose Neural Network that converts the signal data into human pose keypoints. A Multi-Person Tracker then maintains the identities of individuals across frames. The results are delivered via a REST API or WebSocket streaming for real-time applications.

Who it’s for

This tool is designed for developers and organizations working in healthcare (e.g., fall detection), fitness (e.g., form analysis), smart home automation, and security monitoring.

Highlights

• Privacy-First: Detects poses using WiFi signals instead of visual cameras.
• Real-Time Performance: Achieves 30 FPS pose estimation with sub-50ms latency.
• Multi-Person Support: Capable of simultaneously tracking up to 10 individuals.
• Hardware Agnostic: Compatible with various standard WiFi routers and CSI-capable network cards.
• Comprehensive Analytics: Includes built-in capabilities for fall detection, activity recognition, and occupancy monitoring.