WiFi DensePose is a production-grade implementation of WiFi-based dense human pose estimation that enables real-time full-body tracking through walls using commodity mesh routers. With 6,800 GitHub stars and MIT licensing, it demonstrates how Channel State Information from standard WiFi signals can replace cameras for privacy-preserving human pose detection at sub-50ms latency. ## The Case for WiFi-Based Pose Estimation Camera-based pose estimation systems face fundamental limitations. They require line of sight, raise privacy concerns, and struggle in low-light conditions. WiFi signals, by contrast, pass through walls, work in complete darkness, and capture movement data without recording identifiable images. This makes WiFi-based detection inherently more privacy-preserving than any camera system. WiFi DensePose turns this theoretical advantage into a working system. By analyzing the Channel State Information that WiFi routers already collect during normal operation, the system reconstructs dense human body poses without any visual input. ## How It Works ### Channel State Information Processing Every WiFi transmission includes Channel State Information (CSI), which describes how the signal propagates between transmitter and receiver. When a human body moves within the signal path, it causes characteristic distortions in the CSI data. WiFi DensePose uses machine learning models to interpret these distortions and map them to body pose coordinates. The system processes CSI data from standard commodity WiFi routers, meaning no specialized hardware is required. Any mesh router system that exposes CSI data can serve as the sensing infrastructure. ### Neural Network Pipeline The detection pipeline uses a sequence of neural network stages. CSI data is first preprocessed to extract relevant features, then passed through a pose estimation model that outputs dense body coordinates. The pipeline achieves 30 frames per second processing with sub-50ms latency, making it suitable for real-time applications. ### Multi-Person Tracking The system supports simultaneous tracking of up to 10 individuals within the WiFi coverage area. Each person's pose is independently estimated and tracked over time, enabling applications in crowded indoor environments. ## Rust Implementation The v2 release introduced a complete Rust port that delivers dramatic performance improvements over the original Python implementation. CSI preprocessing runs approximately 1,000 times faster in Rust, and the full pipeline achieves an 810x speedup. Throughput reaches approximately 54,000 frames per second capability, while memory usage drops from 500MB to approximately 100MB. The Rust implementation also supports WebAssembly compilation, enabling deployment in browser-based applications. This opens the door to edge computing scenarios where pose estimation runs directly on gateway devices or client hardware. ## Disaster Response Module WiFi-Mat is a specialized disaster response module built on top of WiFi DensePose. Designed for search and rescue operations, it provides vital sign detection through rubble, 3D localization of trapped individuals, and automatic triage classification. The module is specifically designed for earthquake and building collapse scenarios where camera-based systems are useless. This application demonstrates how WiFi-based sensing can save lives in situations where traditional detection methods cannot operate. ## Technical Stack The primary implementation uses Python 3.8 or later with FastAPI as the web framework. Docker support simplifies deployment, and the project is available through PyPI for standard Python installation. Kubernetes deployment files and Terraform infrastructure templates are included for production-scale operations. The REST API includes authentication and rate limiting, reflecting the project's production-grade ambitions. Comprehensive documentation covers setup, configuration, and deployment across different environments. ## Testing and Quality The project claims 100 percent test coverage, which is unusual for a research-oriented machine learning project. This emphasis on testing reflects the production focus and the safety-critical nature of some intended applications, particularly the disaster response module. ## Applications Smart home automation can use WiFi DensePose for presence detection and activity recognition without installing cameras. Healthcare monitoring enables fall detection and movement tracking for elderly care. Security systems can detect unauthorized entry through walls. Fitness applications can track exercise form without camera setup. Retail analytics can measure foot traffic and customer behavior patterns. ## Limitations WiFi-based pose estimation is inherently less precise than camera-based systems. The spatial resolution depends on the number and placement of WiFi access points, and dense body poses may lack the fine-grained detail available from visual systems. The system requires WiFi routers that expose CSI data, which is not universal across all consumer router brands. Environmental factors like metal structures, water pipes, and electronic interference can affect signal quality. The neural network models need training data specific to the deployment environment for optimal accuracy. ## Market Position WiFi DensePose occupies a unique position at the intersection of WiFi sensing, computer vision, and edge computing. The privacy-preserving nature of WiFi-based detection positions it as a compelling alternative to camera systems in settings where visual surveillance raises ethical or legal concerns.