WiFi DensePose

See through walls with WiFi. No cameras. No wearables. Just radio waves.

WiFi DensePose turns commodity WiFi signals into real-time human pose estimation, vital sign monitoring, and presence detection — all without a single pixel of video. By analyzing Channel State Information (CSI) disturbances caused by human movement, the system reconstructs body position, breathing rate, and heartbeat using physics-based signal processing and machine learning.

What How Speed

Pose estimation CSI subcarrier amplitude/phase → DensePose UV maps 54K fps (Rust)

Breathing detection Bandpass 0.1-0.5 Hz → FFT peak 6-30 BPM

Heart rate Bandpass 0.8-2.0 Hz → FFT peak 40-120 BPM

Presence sensing RSSI variance + motion band power < 1ms latency

Through-wall Fresnel zone geometry + multipath modeling Up to 5m depth

# 30 seconds to live sensing — no toolchain required
docker pull ruvnet/wifi-densepose:latest
docker run -p 3000:3000 ruvnet/wifi-densepose:latest
# Open http://localhost:3000

Note

CSI-capable hardware required. Pose estimation, vital signs, and through-wall sensing rely on Channel State Information (CSI) — per-subcarrier amplitude and phase data that standard consumer WiFi does not expose. You need CSI-capable hardware (ESP32-S3 or a research NIC) for full functionality. Consumer WiFi laptops can only provide RSSI-based presence detection, which is significantly less capable.

Hardware options for live CSI capture:

Option Hardware Cost Full CSI Capabilities

ESP32 Mesh (recommended) 3-6x ESP32-S3 + WiFi router ~$54 Yes Pose, breathing, heartbeat, motion, presence

Research NIC Intel 5300 / Atheros AR9580 ~$50-100 Yes Full CSI with 3x3 MIMO

Any WiFi Windows/Linux laptop $0 No RSSI-only: coarse presence and motion

No hardware? Verify the signal processing pipeline with the deterministic reference signal: python v1/data/proof/verify.py

🚀 Key Features

	Feature	What It Means
🔒	Privacy-First	Tracks human pose using only WiFi signals — no cameras, no video, no images stored
⚡	Real-Time	Analyzes WiFi signals in under 100 microseconds per frame — fast enough for live monitoring
💓	Vital Signs	Detects breathing rate (6-30 breaths/min) and heart rate (40-120 bpm) without any wearable
👥	Multi-Person	Tracks multiple people simultaneously, each with independent pose and vitals — no hard software limit (physics: ~3-5 per AP with 56 subcarriers, more with multi-AP)
🧱	Through-Wall	WiFi passes through walls, furniture, and debris — works where cameras cannot
🚑	Disaster Response	Detects trapped survivors through rubble and classifies injury severity (START triage)
🐳	One-Command Setup	`docker pull ruvnet/wifi-densepose:latest` — live sensing in 30 seconds, no toolchain needed
📦	Portable Models	Trained models package into a single `.rvf` file — runs on edge, cloud, or browser (WASM)
🦀	810x Faster	Complete Rust rewrite: 54,000 frames/sec pipeline, 132 MB Docker image, 542+ tests

🏢 Use Cases & Applications

WiFi sensing works anywhere WiFi exists. No new hardware in most cases — just software on existing access points or a $8 ESP32 add-on. Because there are no cameras, deployments avoid privacy regulations (GDPR video, HIPAA imaging) by design.

Scaling: Each AP distinguishes ~3-5 people (56 subcarriers). Multi-AP multiplies linearly — a 4-AP retail mesh covers ~15-20 occupants. No hard software limit; the practical ceiling is signal physics.

	Why WiFi sensing wins	Traditional alternative
🔒	No video, no GDPR/HIPAA imaging rules	Cameras require consent, signage, data retention policies
🧱	Works through walls, shelving, debris	Cameras need line-of-sight per room
🌙	Works in total darkness	Cameras need IR or visible light
💰	$0-$8 per zone (existing WiFi or ESP32)	Camera systems: $200-$2,000 per zone
🔌	WiFi already deployed everywhere	PIR/radar sensors require new wiring per room

🏥 Everyday — Healthcare, retail, office, hospitality (commodity WiFi)

Use Case	What It Does	Hardware	Key Metric
Elderly care / assisted living	Fall detection, nighttime activity monitoring, breathing rate during sleep — no wearable compliance needed	1 ESP32-S3 per room ($8)	Fall alert <2s
Hospital patient monitoring	Continuous breathing + heart rate for non-critical beds without wired sensors; nurse alert on anomaly	1-2 APs per ward	Breathing: 6-30 BPM
Emergency room triage	Automated occupancy count + wait-time estimation; detect patient distress (abnormal breathing) in waiting areas	Existing hospital WiFi	Occupancy accuracy >95%
Retail occupancy & flow	Real-time foot traffic, dwell time by zone, queue length — no cameras, no opt-in, GDPR-friendly	Existing store WiFi + 1 ESP32	Dwell resolution ~1m
Office space utilization	Which desks/rooms are actually occupied, meeting room no-shows, HVAC optimization based on real presence	Existing enterprise WiFi	Presence latency <1s
Hotel & hospitality	Room occupancy without door sensors, minibar/bathroom usage patterns, energy savings on empty rooms	Existing hotel WiFi	15-30% HVAC savings
Restaurants & food service	Table turnover tracking, kitchen staff presence, restroom occupancy displays — no cameras in dining areas	Existing WiFi	Queue wait ±30s
Parking garages	Pedestrian presence in stairwells and elevators where cameras have blind spots; security alert if someone lingers	Existing WiFi	Through-concrete walls

🏟️ Specialized — Events, fitness, education, civic (CSI-capable hardware)

Use Case	What It Does	Hardware	Key Metric
Smart home automation	Room-level presence triggers (lights, HVAC, music) that work through walls — no dead zones, no motion-sensor timeouts	2-3 ESP32-S3 nodes ($24)	Through-wall range ~5m
Fitness & sports	Rep counting, posture correction, breathing cadence during exercise — no wearable, no camera in locker rooms	3+ ESP32-S3 mesh	Pose: 17 keypoints
Childcare & schools	Naptime breathing monitoring, playground headcount, restricted-area alerts — privacy-safe for minors	2-4 ESP32-S3 per zone	Breathing: ±1 BPM
Event venues & concerts	Crowd density mapping, crush-risk detection via breathing compression, emergency evacuation flow tracking	Multi-AP mesh (4-8 APs)	Density per m²
Stadiums & arenas	Section-level occupancy for dynamic pricing, concession staffing, emergency egress flow modeling	Enterprise AP grid	15-20 per AP mesh
Houses of worship	Attendance counting without facial recognition — privacy-sensitive congregations, multi-room campus tracking	Existing WiFi	Zone-level accuracy
Warehouse & logistics	Worker safety zones, forklift proximity alerts, occupancy in hazardous areas — works through shelving and pallets	Industrial AP mesh	Alert latency <500ms
Civic infrastructure	Public restroom occupancy (no cameras possible), subway platform crowding, shelter headcount during emergencies	Municipal WiFi + ESP32	Real-time headcount
Museums & galleries	Visitor flow heatmaps, exhibit dwell time, crowd bottleneck alerts — no cameras near artwork (flash/theft risk)	Existing WiFi	Zone dwell ±5s

🤖 Robotics & Industrial — Autonomous systems, manufacturing, android spatial awareness

WiFi sensing gives robots and autonomous systems a spatial awareness layer that works where LIDAR and cameras fail — through dust, smoke, fog, and around corners. The CSI signal field acts as a "sixth sense" for detecting humans in the environment without requiring line-of-sight.

Use Case	What It Does	Hardware	Key Metric
Cobot safety zones	Detect human presence near collaborative robots — auto-slow or stop before contact, even behind obstructions	2-3 ESP32-S3 per cell	Presence latency <100ms
Warehouse AMR navigation	Autonomous mobile robots sense humans around blind corners, through shelving racks — no LIDAR occlusion	ESP32 mesh along aisles	Through-shelf detection
Android / humanoid spatial awareness	Ambient human pose sensing for social robots — detect gestures, approach direction, and personal space without cameras always on	Onboard ESP32-S3 module	17-keypoint pose
Manufacturing line monitoring	Worker presence at each station, ergonomic posture alerts, headcount for shift compliance — works through equipment	Industrial AP per zone	Pose + breathing
Construction site safety	Exclusion zone enforcement around heavy machinery, fall detection from scaffolding, personnel headcount	Ruggedized ESP32 mesh	Alert <2s, through-dust
Agricultural robotics	Detect farm workers near autonomous harvesters in dusty/foggy field conditions where cameras are unreliable	Weatherproof ESP32 nodes	Range ~10m open field
Drone landing zones	Verify landing area is clear of humans — WiFi sensing works in rain, dust, and low light where downward cameras fail	Ground ESP32 nodes	Presence: >95% accuracy
Clean room monitoring	Personnel tracking without cameras (particle contamination risk from camera fans) — gown compliance via pose	Existing cleanroom WiFi	No particulate emission

🔥 Extreme — Through-wall, disaster, defense, underground

These scenarios exploit WiFi's ability to penetrate solid materials — concrete, rubble, earth — where no optical or infrared sensor can reach. The WiFi-Mat disaster module (ADR-001) is specifically designed for this tier.

Use Case	What It Does	Hardware	Key Metric
Search & rescue (WiFi-Mat)	Detect survivors through rubble/debris via breathing signature, START triage color classification, 3D localization	Portable ESP32 mesh + laptop	Through 30cm concrete
Firefighting	Locate occupants through smoke and walls before entry; breathing detection confirms life signs remotely	Portable mesh on truck	Works in zero visibility
Prison & secure facilities	Cell occupancy verification, distress detection (abnormal vitals), perimeter sensing — no camera blind spots	Dedicated AP infrastructure	24/7 vital signs
Military / tactical	Through-wall personnel detection, room clearing confirmation, hostage vital signs at standoff distance	Directional WiFi + custom FW	Range: 5m through wall
Border & perimeter security	Detect human presence in tunnels, behind fences, in vehicles — passive sensing, no active illumination to reveal position	Concealed ESP32 mesh	Passive / covert
Mining & underground	Worker presence in tunnels where GPS/cameras fail, breathing detection after collapse, headcount at safety points	Ruggedized ESP32 mesh	Through rock/earth
Maritime & naval	Below-deck personnel tracking through steel bulkheads (limited range, requires tuning), man-overboard detection	Ship WiFi + ESP32	Through 1-2 bulkheads
Wildlife research	Non-invasive animal activity monitoring in enclosures or dens — no light pollution, no visual disturbance	Weatherproof ESP32 nodes	Zero light emission

📦 Installation

Guided Installer — Interactive hardware detection and profile selection

./install.sh

The installer walks through 7 steps: system detection, toolchain check, WiFi hardware scan, profile recommendation, dependency install, build, and verification.

Profile	What it installs	Size	Requirements
`verify`	Pipeline verification only	~5 MB	Python 3.8+
`python`	Full Python API server + sensing	~500 MB	Python 3.8+
`rust`	Rust pipeline (~810x faster)	~200 MB	Rust 1.70+
`browser`	WASM for in-browser execution	~10 MB	Rust + wasm-pack
`iot`	ESP32 sensor mesh + aggregator	varies	Rust + ESP-IDF
`docker`	Docker-based deployment	~1 GB	Docker
`field`	WiFi-Mat disaster response kit	~62 MB	Rust + wasm-pack
`full`	Everything available	~2 GB	All toolchains

# Non-interactive
./install.sh --profile rust --yes

# Hardware check only
./install.sh --check-only

From Source — Rust (primary) or Python

git clone https://github.com/ruvnet/wifi-densepose.git
cd wifi-densepose

# Rust (primary — 810x faster)
cd rust-port/wifi-densepose-rs
cargo build --release
cargo test --workspace

# Python (legacy v1)
pip install -r requirements.txt
pip install -e .

# Or via pip
pip install wifi-densepose
pip install wifi-densepose[gpu]   # GPU acceleration
pip install wifi-densepose[all]   # All optional deps

Docker — Pre-built images, no toolchain needed

# Rust sensing server (132 MB — recommended)
docker pull ruvnet/wifi-densepose:latest
docker run -p 3000:3000 -p 3001:3001 -p 5005:5005/udp ruvnet/wifi-densepose:latest

# Python sensing pipeline (569 MB)
docker pull ruvnet/wifi-densepose:python
docker run -p 8765:8765 -p 8080:8080 ruvnet/wifi-densepose:python

# Both via docker-compose
cd docker && docker compose up

# Export RVF model
docker run --rm -v $(pwd):/out ruvnet/wifi-densepose:latest --export-rvf /out/model.rvf

Image	Tag	Size	Ports
`ruvnet/wifi-densepose`	`latest`, `rust`	132 MB	3000 (REST), 3001 (WS), 5005/udp (ESP32)
`ruvnet/wifi-densepose`	`python`	569 MB	8765 (WS), 8080 (UI)

System Requirements

Rust: 1.70+ (primary runtime — install via rustup)
Python: 3.8+ (for verification and legacy v1 API)
OS: Linux (Ubuntu 18.04+), macOS (10.15+), Windows 10+
Memory: Minimum 4GB RAM, Recommended 8GB+
Storage: 2GB free space for models and data
Network: WiFi interface with CSI capability (optional — installer detects what you have)
GPU: Optional (NVIDIA CUDA or Apple Metal)

🚀 Quick Start

First API call in 3 commands

1. Install

# Fastest path — Docker
docker pull ruvnet/wifi-densepose:latest
docker run -p 3000:3000 ruvnet/wifi-densepose:latest

# Or from source (Rust)
./install.sh --profile rust --yes

2. Start the System

from wifi_densepose import WiFiDensePose

system = WiFiDensePose()
system.start()
poses = system.get_latest_poses()
print(f"Detected {len(poses)} persons")
system.stop()

3. REST API

# Health check
curl http://localhost:3000/health

# Latest sensing frame
curl http://localhost:3000/api/v1/sensing/latest

# Vital signs
curl http://localhost:3000/api/v1/vital-signs

# Pose estimation
curl http://localhost:3000/api/v1/pose/current

# Server info
curl http://localhost:3000/api/v1/info

4. Real-time WebSocket

import asyncio, websockets, json

async def stream():
    async with websockets.connect("ws://localhost:3001/ws/sensing") as ws:
        async for msg in ws:
            data = json.loads(msg)
            print(f"Persons: {len(data.get('persons', []))}")

asyncio.run(stream())

📋 Table of Contents

📡 Signal Processing & Sensing — From raw WiFi frames to vital signs

The signal processing stack transforms raw WiFi Channel State Information into actionable human sensing data. Starting from 56-192 subcarrier complex values captured at 20 Hz, the pipeline applies research-grade algorithms (SpotFi phase correction, Hampel outlier rejection, Fresnel zone modeling) to extract breathing rate, heart rate, motion level, and multi-person body pose — all in pure Rust with zero external ML dependencies.

Section	Description	Docs
Key Features	Privacy-first sensing, real-time performance, multi-person tracking, Docker	—
ESP32-S3 Hardware Pipeline	20 Hz CSI streaming, binary frame parsing, flash & provision	ADR-018 · Tutorial #34
Vital Sign Detection	Breathing 6-30 BPM, heartbeat 40-120 BPM, FFT peak detection	ADR-021
WiFi Scan Domain Layer	8-stage RSSI pipeline, multi-BSSID fingerprinting, Windows WiFi	ADR-022 · Tutorial #36
WiFi-Mat Disaster Response	Search & rescue, START triage, 3D localization through debris	ADR-001 · User Guide
SOTA Signal Processing	SpotFi, Hampel, Fresnel, STFT spectrogram, subcarrier selection, BVP	ADR-014

🧠 Models & Training — DensePose pipeline, RVF containers, SONA adaptation, RuVector integration

The neural pipeline uses a graph transformer with cross-attention to map CSI feature matrices to 17 COCO body keypoints and DensePose UV coordinates. Models are packaged as single-file .rvf containers with progressive loading (Layer A instant, Layer B warm, Layer C full). SONA (Self-Optimizing Neural Architecture) enables continuous on-device adaptation via micro-LoRA + EWC++ without catastrophic forgetting. Signal processing is powered by 5 RuVector crates (v2.0.4) with 7 integration points across the Rust workspace, plus 6 additional vendored crates for inference and graph intelligence.

Section	Description	Docs
RVF Model Container	Binary packaging with Ed25519 signing, progressive 3-layer loading, SIMD quantization	ADR-023
Training & Fine-Tuning	8-phase pure Rust pipeline (7,832 lines), MM-Fi/Wi-Pose pre-training, 6-term composite loss, SONA LoRA	ADR-023
RuVector Crates	11 vendored Rust crates from ruvector: attention, min-cut, solver, GNN, HNSW, temporal compression, sparse inference	GitHub · Source

🖥️ Usage & Configuration — CLI flags, API endpoints, hardware setup

The Rust sensing server is the primary interface, offering a comprehensive CLI with flags for data source selection, model loading, training, benchmarking, and RVF export. A REST API (Axum) and WebSocket server provide real-time data access. The Python v1 CLI remains available for legacy workflows.

Section	Description	Docs
CLI Usage	`--source`, `--train`, `--benchmark`, `--export-rvf`, `--model`, `--progressive`	—
REST API & WebSocket	6 REST endpoints (sensing, vitals, BSSID, SONA), WebSocket real-time stream	—
Hardware Support	ESP32-S3 ($8), Intel 5300 ($15), Atheros AR9580 ($20), Windows RSSI ($0)	ADR-012 · ADR-013

⚙️ Development & Testing — 542+ tests, CI, deployment

The project maintains 542+ pure-Rust tests across 7 crate suites with zero mocks — every test runs against real algorithm implementations. Hardware-free simulation mode (--source simulate) enables full-stack testing without physical devices. Docker images are published on Docker Hub for zero-setup deployment.

Section	Description	Docs
Testing	7 test suites: sensing-server (229), signal (83), mat (139), wifiscan (91), RVF (16), vitals (18)	—
Deployment	Docker images (132 MB Rust / 569 MB Python), docker-compose, env vars	—
Contributing	Fork → branch → test → PR workflow, Rust and Python dev setup	—

📊 Performance & Benchmarks — Measured throughput, latency, resource usage

All benchmarks are measured on the Rust sensing server using cargo bench and the built-in --benchmark CLI flag. The Rust v2 implementation delivers 810x end-to-end speedup over the Python v1 baseline, with motion detection reaching 5,400x improvement. The vital sign detector processes 11,665 frames/second in a single-threaded benchmark.

Section	Description	Key Metric
Performance Metrics	Vital signs, CSI pipeline, motion detection, Docker image, memory	11,665 fps vitals · 54K fps pipeline
Rust vs Python	Side-by-side benchmarks across 5 operations	810x full pipeline speedup

📄 Meta — License, changelog, support

WiFi DensePose is MIT-licensed open source, developed by ruvnet. The project has been in active development since March 2025, with 3 major releases delivering the Rust port, SOTA signal processing, disaster response module, and end-to-end training pipeline.

Section	Description	Link
Changelog	v2.3.0 (training pipeline + Docker), v2.2.0 (SOTA + WiFi-Mat), v2.1.0 (Rust port)	—
License	MIT License	LICENSE
Support	Bug reports, feature requests, community discussion	Issues · Discussions

📡 Signal Processing & Sensing

📡 ESP32-S3 Hardware Pipeline (ADR-018) — 20 Hz CSI streaming, flash & provision

ESP32-S3 (STA + promiscuous)     UDP/5005      Rust aggregator
┌─────────────────────────┐    ──────────>    ┌──────────────────┐
│ WiFi CSI callback 20 Hz │    ADR-018        │ Esp32CsiParser   │
│ ADR-018 binary frames   │    binary         │ CsiFrame output  │
│ stream_sender (UDP)     │                   │ presence detect  │
└─────────────────────────┘                   └──────────────────┘

Metric	Measured
Frame rate	~20 Hz sustained
Subcarriers	64 / 128 / 192 (LLTF, HT, HT40)
Latency	< 1ms (UDP loopback)
Presence detection	Motion score 10/10 at 3m

# Pre-built binaries — no toolchain required
# https://github.com/ruvnet/wifi-densepose/releases/tag/v0.1.0-esp32

python -m esptool --chip esp32s3 --port COM7 --baud 460800 \
  write-flash --flash-mode dio --flash-size 4MB \
  0x0 bootloader.bin 0x8000 partition-table.bin 0x10000 esp32-csi-node.bin

python scripts/provision.py --port COM7 \
  --ssid "YourWiFi" --password "secret" --target-ip 192.168.1.20

cargo run -p wifi-densepose-hardware --bin aggregator -- --bind 0.0.0.0:5005 --verbose

See firmware/esp32-csi-node/README.md and Tutorial #34.

🦀 Rust Implementation (v2) — 810x faster, 54K fps pipeline

Performance Benchmarks (Validated)

Operation	Python (v1)	Rust (v2)	Speedup
CSI Preprocessing (4x64)	~5ms	5.19 µs	~1000x
Phase Sanitization (4x64)	~3ms	3.84 µs	~780x
Feature Extraction (4x64)	~8ms	9.03 µs	~890x
Motion Detection	~1ms	186 ns	~5400x
Full Pipeline	~15ms	18.47 µs	~810x
Vital Signs	N/A	86 µs	11,665 fps

Resource	Python (v1)	Rust (v2)
Memory	~500 MB	~100 MB
Docker Image	569 MB	132 MB
Tests	41	542+
WASM Support	No	Yes

cd rust-port/wifi-densepose-rs
cargo build --release
cargo test --workspace
cargo bench --package wifi-densepose-signal

💓 Vital Sign Detection (ADR-021) — Breathing and heartbeat via FFT

Capability	Range	Method
Breathing Rate	6-30 BPM (0.1-0.5 Hz)	Bandpass filter + FFT peak detection
Heart Rate	40-120 BPM (0.8-2.0 Hz)	Bandpass filter + FFT peak detection
Sampling Rate	20 Hz (ESP32 CSI)	Real-time streaming
Confidence	0.0-1.0 per sign	Spectral coherence + signal quality

./target/release/sensing-server --source simulate --ui-path ../../ui
curl http://localhost:8080/api/v1/vital-signs

See ADR-021.

📡 WiFi Scan Domain Layer (ADR-022) — 8-stage RSSI pipeline for Windows WiFi

Stage	Purpose
Predictive Gating	Pre-filter scan results using temporal prediction
Attention Weighting	Weight BSSIDs by signal relevance
Spatial Correlation	Cross-AP spatial signal correlation
Motion Estimation	Detect movement from RSSI variance
Breathing Extraction	Extract respiratory rate from sub-Hz oscillations
Quality Gating	Reject low-confidence estimates
Fingerprint Matching	Location and posture classification via RF fingerprints
Orchestration	Fuse all stages into unified sensing output

cargo test -p wifi-densepose-wifiscan

See ADR-022 and Tutorial #36.

🚨 WiFi-Mat: Disaster Response — Search & rescue, START triage, 3D localization

WiFi signals penetrate non-metallic debris (concrete, wood, drywall) where cameras and thermal sensors cannot reach. The WiFi-Mat module (wifi-densepose-mat, 139 tests) uses CSI analysis to detect survivors trapped under rubble, classify their condition using the START triage protocol, and estimate their 3D position — giving rescue teams actionable intelligence within seconds of deployment.

Capability	How It Works	Performance Target
Breathing Detection	Bandpass 0.07-1.0 Hz + Fresnel zone modeling detects chest displacement of 5-10mm at 5 GHz	4-60 BPM, <500ms latency
Heartbeat Detection	Micro-Doppler shift extraction from fine-grained CSI phase variation	Via ruvector-temporal-tensor
3D Localization	Multi-AP triangulation + CSI fingerprint matching + depth estimation through rubble layers	3-5m penetration
START Triage	Ensemble classifier votes on breathing + movement + vital stability → P1-P4 priority	<1% false negative
Zone Scanning	16+ concurrent scan zones with periodic re-scan and audit logging	Full disaster site

Triage classification (START protocol compatible):

Status	Color	Detection Criteria	Priority
Immediate	Red	Breathing detected, no movement	P1
Delayed	Yellow	Movement + breathing, stable vitals	P2
Minor	Green	Strong movement, responsive patterns	P3
Deceased	Black	No vitals for >30 min continuous scan	P4

Deployment modes: portable (single TX/RX handheld), distributed (multiple APs around collapse site), drone-mounted (UAV scanning), vehicle-mounted (mobile command post).

use wifi_densepose_mat::{DisasterResponse, DisasterConfig, DisasterType, ScanZone, ZoneBounds};

let config = DisasterConfig::builder()
    .disaster_type(DisasterType::Earthquake)
    .sensitivity(0.85)
    .max_depth(5.0)
    .build();

let mut response = DisasterResponse::new(config);
response.initialize_event(location, "Building collapse")?;
response.add_zone(ScanZone::new("North Wing", ZoneBounds::rectangle(0.0, 0.0, 30.0, 20.0)))?;
response.start_scanning().await?;

Safety guarantees: fail-safe defaults (assume life present on ambiguous signals), redundant multi-algorithm voting, complete audit trail, offline-capable (no network required).

WiFi-Mat User Guide | ADR-001 | Domain Model

🔬 SOTA Signal Processing (ADR-014) — 6 research-grade algorithms

The signal processing layer bridges the gap between raw commodity WiFi hardware output and research-grade sensing accuracy. Each algorithm addresses a specific limitation of naive CSI processing — from hardware-induced phase corruption to environment-dependent multipath interference. All six are implemented in wifi-densepose-signal/src/ with deterministic tests and no mock data.

Algorithm	What It Does	Why It Matters	Math	Source
Conjugate Multiplication	Multiplies CSI antenna pairs: `H₁[k] × conj(H₂[k])`	Cancels CFO, SFO, and packet detection delay that corrupt raw phase — preserves only environment-caused phase differences	`CSI_ratio[k] = H₁[k] * conj(H₂[k])`	SpotFi (SIGCOMM 2015)
Hampel Filter	Replaces outliers using running median ± scaled MAD	Z-score uses mean/std which are corrupted by the very outliers it detects (masking effect). Hampel uses median/MAD, resisting up to 50% contamination	`σ̂ = 1.4826 × MAD`	Standard DSP; WiGest (2015)
Fresnel Zone Model	Models signal variation from chest displacement crossing Fresnel zone boundaries	Zero-crossing counting fails in multipath-rich environments. Fresnel predicts where breathing should appear based on TX-RX-body geometry	`ΔΦ = 2π × 2Δd / λ`, `A = \|sin(ΔΦ/2)\|`	FarSense (MobiCom 2019)
CSI Spectrogram	Sliding-window FFT (STFT) per subcarrier → 2D time-frequency matrix	Breathing = 0.2-0.4 Hz band, walking = 1-2 Hz, static = noise. 2D structure enables CNN spatial pattern recognition that 1D features miss	`S[t,f] = \|Σₙ x[n] w[n-t] e^{-j2πfn}\|²`	Standard since 2018
Subcarrier Selection	Ranks subcarriers by motion sensitivity (variance ratio) and selects top-K	Not all subcarriers respond to motion — some sit in multipath nulls. Selecting the 10-20 most sensitive improves SNR by 6-10 dB	`sensitivity[k] = var_motion / var_static`	WiDance (MobiCom 2017)
Body Velocity Profile	Extracts velocity distribution from Doppler shifts across subcarriers	BVP is domain-independent — same velocity profile regardless of room layout, furniture, or AP placement. Basis for cross-environment recognition	`BVP[v,t] = Σₖ \|STFTₖ[v,t]\|`	Widar 3.0 (MobiSys 2019)

Processing pipeline order: Raw CSI → Conjugate multiplication (phase cleaning) → Hampel filter (outlier removal) → Subcarrier selection (top-K) → CSI spectrogram (time-frequency) → Fresnel model (breathing) + BVP (activity)

See ADR-014 for full mathematical derivations.

🧠 Models & Training

📦 RVF Model Container — Single-file deployment with progressive loading

The RuVector Format (RVF) packages an entire trained model — weights, HNSW indexes, quantization codebooks, SONA adaptation deltas, and WASM inference runtime — into a single self-contained binary file. No external dependencies are needed at deployment time.

Container structure:

┌──────────────────────────────────────────────────────┐
│ RVF Container (.rvf)                                  │
│                                                       │
│  ┌─────────────┐  64-byte header per segment          │
│  │ Manifest     │  Magic: 0x52564653 ("RVFS")         │
│  ├─────────────┤  Type + content hash + compression   │
│  │ Weights      │  Model parameters (f32/f16/u8)      │
│  ├─────────────┤                                      │
│  │ HNSW Index   │  Vector search index                │
│  ├─────────────┤                                      │
│  │ Quant        │  Quantization codebooks              │
│  ├─────────────┤                                      │
│  │ SONA Profile │  LoRA deltas + EWC++ Fisher matrix  │
│  ├─────────────┤                                      │
│  │ Witness      │  Ed25519 training proof              │
│  ├─────────────┤                                      │
│  │ Vitals Config│  Breathing/HR filter parameters     │
│  └─────────────┘                                      │
└──────────────────────────────────────────────────────┘

Deployment targets:

Target	Quantization	Size	Load Time	Use Case
ESP32 / IoT	int4	~0.7 MB	<5ms (Layer A)	Presence + breathing only
Mobile / WebView	int8	~6 MB	~200ms (Layer B)	Pose estimation on phone
Browser (WASM)	int8	~10 MB	~500ms (Layer B)	In-browser demo
Field (WiFi-Mat)	fp16	~62 MB	~2s (Layer C)	Full DensePose + disaster triage
Server / Cloud	f32	~50+ MB	~3s (Layer C)	Training + full inference

Property	Detail
Format	Segment-based binary, 20+ segment types, CRC32 integrity per segment
Progressive Loading	Layer A (<5ms): manifest + entry points → Layer B (100ms-1s): hot weights + adjacency → Layer C (seconds): full graph
Signing	Ed25519 training proofs for verifiable provenance — chain of custody from training data to deployed model
Quantization	Per-segment temperature-tiered: f32 (full), f16 (half), u8 (int8), int4 — with SIMD-accelerated distance computation
CLI	`--export-rvf` (generate), `--load-rvf` (config), `--save-rvf` (persist), `--model` (inference), `--progressive` (3-layer load)

# Export model package
./target/release/sensing-server --export-rvf wifi-densepose-v1.rvf

# Load and run with progressive loading
./target/release/sensing-server --model wifi-densepose-v1.rvf --progressive

# Export via Docker
docker run --rm -v $(pwd):/out ruvnet/wifi-densepose:latest --export-rvf /out/model.rvf

Built on the rvf crate family (rvf-types, rvf-wire, rvf-manifest, rvf-index, rvf-quant, rvf-crypto, rvf-runtime). See ADR-023.

🧬 Training & Fine-Tuning — MM-Fi/Wi-Pose pre-training, SONA adaptation

The training pipeline implements 8 phases in pure Rust (7,832 lines, zero external ML dependencies). It trains a graph transformer with cross-attention to map CSI feature matrices to 17 COCO body keypoints and DensePose UV coordinates — following the approach from the CMU "DensePose From WiFi" paper (arXiv:2301.00250). RuVector crates provide the core building blocks: ruvector-attention for cross-attention layers, ruvector-mincut for multi-person matching, and ruvector-temporal-tensor for CSI buffer compression.

Three-tier data strategy:

Tier	Method	Purpose	RuVector Integration
1. Pre-train	MM-Fi + Wi-Pose public datasets	Cross-environment generalization (multi-subject, multi-room)	`ruvector-temporal-tensor` compresses CSI windows (114→56 subcarrier resampling)
2. Fine-tune	ESP32 CSI + camera pseudo-labels	Environment-specific multipath adaptation	`ruvector-solver` for Fresnel geometry, `ruvector-attn-mincut` for subcarrier gating
3. SONA adapt	Micro-LoRA (rank-4) + EWC++	Continuous on-device learning without catastrophic forgetting	SONA architecture (Self-Optimizing Neural Architecture)

Training pipeline components:

Phase	Module	What It Does	RuVector Crate
1	`dataset.rs` (850 lines)	MM-Fi `.npy` + Wi-Pose `.mat` loaders, subcarrier resampling (114→56, 30→56), windowing	ruvector-temporal-tensor
2	`graph_transformer.rs` (855 lines)	COCO BodyGraph (17 kp, 16 edges), AntennaGraph, multi-head CrossAttention, GCN message passing	ruvector-attention
3	`trainer.rs` (881 lines)	6-term composite loss (MSE, CE, UV, temporal, bone, symmetry), SGD+momentum, cosine+warmup, PCK/OKS	ruvector-mincut (person matching)
4	`sona.rs` (639 lines)	LoRA adapters (A×B delta), EWC++ Fisher regularization, EnvironmentDetector (3-sigma drift)	sona
5	`sparse_inference.rs` (753 lines)	NeuronProfiler hot/cold partitioning, SparseLinear (skip cold rows), INT8/FP16 quantization	ruvector-sparse-inference
6	`rvf_pipeline.rs` (1,027 lines)	Progressive 3-layer loader, HNSW index, OverlayGraph, `RvfModelBuilder`	ruvector-core (HNSW)
7	`rvf_container.rs` (914 lines)	Binary container format, 6+ segment types, CRC32 integrity	rvf
8	`main.rs` integration	`--train`, `--model`, `--progressive` CLI flags, REST endpoints	—

SONA (Self-Optimizing Neural Architecture) — the continuous adaptation system:

Component	What It Does	Why It Matters
Micro-LoRA (rank-4)	Trains small A×B weight deltas instead of full weights	100x fewer parameters to update → runs on ESP32
EWC++ (Fisher matrix)	Penalizes changes to important weights from previous environments	Prevents catastrophic forgetting when moving between rooms
EnvironmentDetector	Monitors CSI feature drift with 3-sigma threshold	Auto-triggers adaptation when the model is moved to a new space
Best-epoch snapshot	Saves best validation loss weights, restores before export	Prevents shipping overfit final-epoch parameters

# Pre-train on MM-Fi dataset
./target/release/sensing-server --train --dataset data/ --dataset-type mmfi --epochs 100

# Train and export to RVF in one step
./target/release/sensing-server --train --dataset data/ --epochs 100 --save-rvf model.rvf

# Via Docker (no toolchain needed)
docker run --rm -v $(pwd)/data:/data ruvnet/wifi-densepose:latest \
  --train --dataset /data --epochs 100 --export-rvf /data/model.rvf

See ADR-023 · SONA crate · arXiv:2301.00250

🔩 RuVector Crates — 11 vendored signal intelligence crates from github.com/ruvnet/ruvector

5 directly-used crates (v2.0.4, declared in Cargo.toml, 7 integration points):

Crate	What It Does	Where It's Used in WiFi-DensePose	Source
`ruvector-attention`	Scaled dot-product attention, MoE routing, sparse attention	`model.rs` (spatial attention), `bvp.rs` (sensitivity-weighted velocity profiles)	crate
`ruvector-mincut`	Subpolynomial dynamic min-cut O(n^1.5 log n)	`metrics.rs` (DynamicPersonMatcher — multi-person assignment), `subcarrier_selection.rs` (sensitive/insensitive split)	crate
`ruvector-attn-mincut`	Attention-gated spectrogram noise suppression	`model.rs` (antenna attention gating), `spectrogram.rs` (gate noisy time-frequency bins)	crate
`ruvector-solver`	Sparse Neumann series solver O(sqrt(n))	`fresnel.rs` (TX-body-RX geometry), `triangulation.rs` (3D localization), `subcarrier.rs` (sparse interpolation 114→56)	crate
`ruvector-temporal-tensor`	Tiered temporal compression (8/7/5/3-bit)	`dataset.rs` (CSI buffer compression), `breathing.rs` + `heartbeat.rs` (compressed vital sign spectrograms)	crate

6 additional vendored crates (used by training pipeline and inference):

Crate	What It Does	Source
`ruvector-core`	VectorDB engine, HNSW index, SIMD distance functions, quantization codebooks	crate
`ruvector-gnn`	Graph neural network layers, graph attention, EWC-regularized training	crate
`ruvector-graph-transformer`	Proof-gated graph transformer with cross-attention	crate
`ruvector-sparse-inference`	PowerInfer-style hot/cold neuron partitioning, skip cold rows at runtime	crate
`ruvector-nervous-system`	PredictiveLayer, OscillatoryRouter, Hopfield associative memory	crate
`ruvector-coherence`	Spectral coherence monitoring, HNSW graph health, Fiedler connectivity	crate

The full RuVector ecosystem includes 90+ crates. See github.com/ruvnet/ruvector for the complete library, and vendor/ruvector/ for the vendored source in this project.

🏗️ System Architecture — End-to-end data flow from CSI capture to REST/WebSocket API

End-to-End Pipeline

graph TB
    subgraph HW ["📡 Hardware Layer"]
        direction LR
        R1["WiFi Router 1<br/><small>CSI Source</small>"]
        R2["WiFi Router 2<br/><small>CSI Source</small>"]
        R3["WiFi Router 3<br/><small>CSI Source</small>"]
        ESP["ESP32-S3 Mesh<br/><small>20 Hz · 56 subcarriers</small>"]
        WIN["Windows WiFi<br/><small>RSSI scanning</small>"]
    end

    subgraph INGEST ["⚡ Ingestion"]
        AGG["Aggregator<br/><small>UDP :5005 · ADR-018 frames</small>"]
        BRIDGE["Bridge<br/><small>I/Q → amplitude + phase</small>"]
    end

    subgraph SIGNAL ["🔬 Signal Processing — RuVector v2.0.4"]
        direction TB
        PHASE["Phase Sanitization<br/><small>SpotFi conjugate multiply</small>"]
        HAMPEL["Hampel Filter<br/><small>Outlier rejection · σ=3</small>"]
        SUBSEL["Subcarrier Selection<br/><small>ruvector-mincut · sensitive/insensitive split</small>"]
        SPEC["Spectrogram<br/><small>ruvector-attn-mincut · gated STFT</small>"]
        FRESNEL["Fresnel Geometry<br/><small>ruvector-solver · TX-body-RX distance</small>"]
        BVP["Body Velocity Profile<br/><small>ruvector-attention · weighted BVP</small>"]
    end

    subgraph ML ["🧠 Neural Pipeline"]
        direction TB
        GRAPH["Graph Transformer<br/><small>17 COCO keypoints · 16 edges</small>"]
        CROSS["Cross-Attention<br/><small>CSI features → body pose</small>"]
        SONA["SONA Adapter<br/><small>LoRA rank-4 · EWC++</small>"]
    end

    subgraph VITAL ["💓 Vital Signs"]
        direction LR
        BREATH["Breathing<br/><small>0.1–0.5 Hz · FFT peak</small>"]
        HEART["Heart Rate<br/><small>0.8–2.0 Hz · FFT peak</small>"]
        MOTION["Motion Level<br/><small>Variance + band power</small>"]
    end

    subgraph API ["🌐 Output Layer"]
        direction LR
        REST["REST API<br/><small>Axum :3000 · 6 endpoints</small>"]
        WS["WebSocket<br/><small>:3001 · real-time stream</small>"]
        ANALYTICS["Analytics<br/><small>Fall · Activity · START triage</small>"]
        UI["Web UI<br/><small>Three.js · Gaussian splats</small>"]
    end

    R1 & R2 & R3 --> AGG
    ESP --> AGG
    WIN --> BRIDGE
    AGG --> BRIDGE
    BRIDGE --> PHASE
    PHASE --> HAMPEL
    HAMPEL --> SUBSEL
    SUBSEL --> SPEC
    SPEC --> FRESNEL
    FRESNEL --> BVP
    BVP --> GRAPH
    GRAPH --> CROSS
    CROSS --> SONA
    SONA --> BREATH & HEART & MOTION
    BREATH & HEART & MOTION --> REST & WS & ANALYTICS
    WS --> UI

    style HW fill:#1a1a2e,stroke:#e94560,color:#eee
    style INGEST fill:#16213e,stroke:#0f3460,color:#eee
    style SIGNAL fill:#0f3460,stroke:#533483,color:#eee
    style ML fill:#533483,stroke:#e94560,color:#eee
    style VITAL fill:#2d132c,stroke:#e94560,color:#eee
    style API fill:#1a1a2e,stroke:#0f3460,color:#eee

Signal Processing Detail

graph LR
    subgraph RAW ["Raw CSI Frame"]
        IQ["I/Q Samples<br/><small>56–192 subcarriers × N antennas</small>"]
    end

    subgraph CLEAN ["Phase Cleanup"]
        CONJ["Conjugate Multiply<br/><small>Remove carrier freq offset</small>"]
        UNWRAP["Phase Unwrap<br/><small>Remove 2π discontinuities</small>"]
        HAMPEL2["Hampel Filter<br/><small>Remove impulse noise</small>"]
    end

    subgraph SELECT ["Subcarrier Intelligence"]
        MINCUT["Min-Cut Partition<br/><small>ruvector-mincut</small>"]
        GATE["Attention Gate<br/><small>ruvector-attn-mincut</small>"]
    end

    subgraph EXTRACT ["Feature Extraction"]
        STFT["STFT Spectrogram<br/><small>Time-frequency decomposition</small>"]
        FRESNELZ["Fresnel Zones<br/><small>ruvector-solver</small>"]
        BVPE["BVP Estimation<br/><small>ruvector-attention</small>"]
    end

    subgraph OUT ["Output Features"]
        AMP["Amplitude Matrix"]
        PHASE2["Phase Matrix"]
        DOPPLER["Doppler Shifts"]
        VITALS["Vital Band Power"]
    end

    IQ --> CONJ --> UNWRAP --> HAMPEL2
    HAMPEL2 --> MINCUT --> GATE
    GATE --> STFT --> FRESNELZ --> BVPE
    BVPE --> AMP & PHASE2 & DOPPLER & VITALS

    style RAW fill:#0d1117,stroke:#58a6ff,color:#c9d1d9
    style CLEAN fill:#161b22,stroke:#58a6ff,color:#c9d1d9
    style SELECT fill:#161b22,stroke:#d29922,color:#c9d1d9
    style EXTRACT fill:#161b22,stroke:#3fb950,color:#c9d1d9
    style OUT fill:#0d1117,stroke:#8b949e,color:#c9d1d9

Deployment Topology

graph TB
    subgraph EDGE ["Edge (ESP32-S3 Mesh)"]
        E1["Node 1<br/><small>Kitchen</small>"]
        E2["Node 2<br/><small>Living room</small>"]
        E3["Node 3<br/><small>Bedroom</small>"]
    end

    subgraph SERVER ["Server (Rust · 132 MB Docker)"]
        SENSE["Sensing Server<br/><small>:3000 REST · :3001 WS · :5005 UDP</small>"]
        RVF["RVF Model<br/><small>Progressive 3-layer load</small>"]
        STORE["Time-Series Store<br/><small>In-memory ring buffer</small>"]
    end

    subgraph CLIENT ["Clients"]
        BROWSER["Browser<br/><small>Three.js UI · Gaussian splats</small>"]
        MOBILE["Mobile App<br/><small>WebSocket stream</small>"]
        DASH["Dashboard<br/><small>REST polling</small>"]
        IOT["Home Automation<br/><small>MQTT bridge</small>"]
    end

    E1 -->|"UDP :5005<br/>ADR-018 frames"| SENSE
    E2 -->|"UDP :5005"| SENSE
    E3 -->|"UDP :5005"| SENSE
    SENSE <--> RVF
    SENSE <--> STORE
    SENSE -->|"WS :3001<br/>real-time JSON"| BROWSER & MOBILE
    SENSE -->|"REST :3000<br/>on-demand"| DASH & IOT

    style EDGE fill:#1a1a2e,stroke:#e94560,color:#eee
    style SERVER fill:#16213e,stroke:#533483,color:#eee
    style CLIENT fill:#0f3460,stroke:#0f3460,color:#eee

Component	Crate / Module	Description
Aggregator	`wifi-densepose-hardware`	ESP32 UDP listener, ADR-018 frame parser, I/Q → amplitude/phase bridge
Signal Processor	`wifi-densepose-signal`	SpotFi phase sanitization, Hampel filter, STFT spectrogram, Fresnel geometry, BVP
Subcarrier Selection	`ruvector-mincut` + `ruvector-attn-mincut`	Dynamic sensitive/insensitive partitioning, attention-gated noise suppression
Fresnel Solver	`ruvector-solver`	Sparse Neumann series O(sqrt(n)) for TX-body-RX distance estimation
Graph Transformer	`wifi-densepose-train`	COCO BodyGraph (17 kp, 16 edges), cross-attention CSI→pose, GCN message passing
SONA	`sona` crate	Micro-LoRA (rank-4) adaptation, EWC++ catastrophic forgetting prevention
Vital Signs	`wifi-densepose-signal`	FFT-based breathing (0.1-0.5 Hz) and heartbeat (0.8-2.0 Hz) extraction
REST API	`wifi-densepose-sensing-server`	Axum server: `/api/v1/sensing`, `/health`, `/vital-signs`, `/bssid`, `/sona`
WebSocket	`wifi-densepose-sensing-server`	Real-time pose, sensing, and vital sign streaming on `:3001`
Analytics	`wifi-densepose-mat`	Fall detection, activity recognition, START triage (WiFi-Mat disaster module)
Web UI	`ui/`	Three.js scene, Gaussian splat visualization, signal dashboard

🖥️ CLI Usage

Rust Sensing Server — Primary CLI interface

# Start with simulated data (no hardware)
./target/release/sensing-server --source simulate --ui-path ../../ui

# Start with ESP32 CSI hardware
./target/release/sensing-server --source esp32 --udp-port 5005

# Start with Windows WiFi RSSI
./target/release/sensing-server --source wifi

# Run vital sign benchmark
./target/release/sensing-server --benchmark

# Export RVF model package
./target/release/sensing-server --export-rvf model.rvf

# Train a model
./target/release/sensing-server --train --dataset data/ --epochs 100

# Load trained model with progressive loading
./target/release/sensing-server --model wifi-densepose-v1.rvf --progressive

Flag	Description
`--source`	Data source: `auto`, `wifi`, `esp32`, `simulate`
`--http-port`	HTTP port for UI and REST API (default: 8080)
`--ws-port`	WebSocket port (default: 8765)
`--udp-port`	UDP port for ESP32 CSI frames (default: 5005)
`--benchmark`	Run vital sign benchmark (1000 frames) and exit
`--export-rvf`	Export RVF container package and exit
`--load-rvf`	Load model config from RVF container
`--save-rvf`	Save model state on shutdown
`--model`	Load trained `.rvf` model for inference
`--progressive`	Enable progressive loading (Layer A instant start)
`--train`	Train a model and exit
`--dataset`	Path to dataset directory (MM-Fi or Wi-Pose)
`--epochs`	Training epochs (default: 100)

REST API & WebSocket — Endpoints reference

REST API (Rust Sensing Server)

GET  /api/v1/sensing              # Latest sensing frame
GET  /api/v1/vital-signs          # Breathing, heart rate, confidence
GET  /api/v1/bssid                # Multi-BSSID registry
GET  /api/v1/model/layers         # Progressive loading status
GET  /api/v1/model/sona/profiles  # SONA profiles
POST /api/v1/model/sona/activate  # Activate SONA profile

WebSocket: ws://localhost:8765/ws/sensing (real-time sensing + vital signs)

Default ports: HTTP 8080, WS 8765. Docker images remap to 3000/3001 via --http-port / --ws-port.

Hardware Support — Devices, cost, and guides

Hardware	CSI	Cost	Guide
ESP32-S3	Native	~$8	Tutorial #34
Intel 5300	Firmware mod	~$15	Linux `iwl-csi`
Atheros AR9580	ath9k patch	~$20	Linux only
Any Windows WiFi	RSSI only	$0	Tutorial #36

Python Legacy CLI — v1 API server commands

wifi-densepose start                    # Start API server
wifi-densepose -c config.yaml start     # Custom config
wifi-densepose -v start                 # Verbose logging
wifi-densepose status                   # Check status
wifi-densepose stop                     # Stop server
wifi-densepose config show              # Show configuration
wifi-densepose db init                  # Initialize database
wifi-densepose tasks list               # List background tasks

Documentation Links

WiFi-Mat User Guide | Domain Model
ADR-021 | ADR-022 | ADR-023

🧪 Testing

542+ tests across 7 suites — zero mocks, hardware-free simulation

# Rust tests (primary — 542+ tests)
cd rust-port/wifi-densepose-rs
cargo test --workspace

# Sensing server tests (229 tests)
cargo test -p wifi-densepose-sensing-server

# Vital sign benchmark
./target/release/sensing-server --benchmark

# Python tests
python -m pytest v1/tests/ -v

# Pipeline verification (no hardware needed)
./verify

Suite	Tests	What It Covers
sensing-server lib	147	Graph transformer, trainer, SONA, sparse inference, RVF
sensing-server bin	48	CLI integration, WebSocket, REST API
RVF integration	16	Container build, read, progressive load
Vital signs integration	18	FFT detection, breathing, heartbeat
wifi-densepose-signal	83	SOTA algorithms, Doppler, Fresnel
wifi-densepose-mat	139	Disaster response, triage, localization
wifi-densepose-wifiscan	91	8-stage RSSI pipeline

🚀 Deployment

Docker deployment — Production setup with docker-compose

# Rust sensing server (132 MB)
docker pull ruvnet/wifi-densepose:latest
docker run -p 3000:3000 -p 3001:3001 -p 5005:5005/udp ruvnet/wifi-densepose:latest

# Python pipeline (569 MB)
docker pull ruvnet/wifi-densepose:python
docker run -p 8765:8765 -p 8080:8080 ruvnet/wifi-densepose:python

# Both via docker-compose
cd docker && docker compose up

# Export RVF model
docker run --rm -v $(pwd):/out ruvnet/wifi-densepose:latest --export-rvf /out/model.rvf

Environment Variables

RUST_LOG=info                    # Logging level
WIFI_INTERFACE=wlan0             # WiFi interface for RSSI
POSE_CONFIDENCE_THRESHOLD=0.7    # Minimum confidence
POSE_MAX_PERSONS=10              # Max tracked individuals

📊 Performance Metrics

Measured benchmarks — Rust sensing server, validated via cargo bench

Rust Sensing Server

Metric	Value
Vital sign detection	11,665 fps (86 µs/frame)
Full CSI pipeline	54,000 fps (18.47 µs/frame)
Motion detection	186 ns (~5,400x vs Python)
Docker image	132 MB
Memory usage	~100 MB
Test count	542+

Python vs Rust

Operation	Python	Rust	Speedup
CSI Preprocessing	~5 ms	5.19 µs	1000x
Phase Sanitization	~3 ms	3.84 µs	780x
Feature Extraction	~8 ms	9.03 µs	890x
Motion Detection	~1 ms	186 ns	5400x
Full Pipeline	~15 ms	18.47 µs	810x

🤝 Contributing

Dev setup, code standards, PR process

git clone https://github.com/ruvnet/wifi-densepose.git
cd wifi-densepose

# Rust development
cd rust-port/wifi-densepose-rs
cargo build --release
cargo test --workspace

# Python development
python -m venv venv && source venv/bin/activate
pip install -r requirements-dev.txt && pip install -e .
pre-commit install

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Commit your changes
Push and open a Pull Request

📄 Changelog

Release history

v2.3.0 — 2026-03-01

The largest release to date — delivers the complete end-to-end training pipeline, Docker images, and vital sign detection. The Rust sensing server now supports full model training, RVF export, and progressive model loading from a single binary.

Docker images published — ruvnet/wifi-densepose:latest (132 MB Rust) and :python (569 MB)
8-phase DensePose training pipeline (ADR-023) — Dataset loaders (MM-Fi, Wi-Pose), graph transformer with cross-attention, 6-term composite loss, cosine-scheduled SGD, PCK/OKS validation, SONA adaptation, sparse inference engine, RVF model packaging
--export-rvf CLI flag — Standalone RVF model container generation with vital config, training proof, and SONA profiles
--train CLI flag — Full training mode with best-epoch snapshotting and checkpoint saving
Vital sign detection (ADR-021) — FFT-based breathing (6-30 BPM) and heartbeat (40-120 BPM) extraction, 11,665 fps benchmark
WiFi scan domain layer (ADR-022) — 8-stage pure-Rust signal intelligence pipeline for Windows WiFi RSSI
New crates — wifi-densepose-vitals (1,863 lines) and wifi-densepose-wifiscan (4,829 lines)
542+ Rust tests — All passing, zero mocks

v2.2.0 — 2026-02-28

Introduced the guided installer, SOTA signal processing algorithms, and the WiFi-Mat disaster response module. This release established the ESP32 hardware path and security hardening.

Guided installer — ./install.sh with 7-step hardware detection and 8 install profiles
6 SOTA signal algorithms (ADR-014) — SpotFi conjugate multiplication, Hampel filter, Fresnel zone model, CSI spectrogram, subcarrier selection, body velocity profile
WiFi-Mat disaster response — START triage, scan zones, 3D localization, priority alerts — 139 tests
ESP32 CSI hardware parser — Binary frame parsing with I/Q extraction — 28 tests
Security hardening — 10 vulnerabilities fixed (CVE remediation, input validation, path security)

v2.1.0 — 2026-02-28

The foundational Rust release — ported the Python v1 pipeline to Rust with 810x speedup, integrated the RuVector signal intelligence crates, and added the Three.js real-time visualization.

RuVector integration — 11 vendored crates (ADR-002 through ADR-013) for HNSW indexing, attention, GNN, temporal compression, min-cut, solver
ESP32 CSI sensor mesh — $54 starter kit with 3-6 ESP32-S3 nodes streaming at 20 Hz
Three.js visualization — 3D body model with 17 joints, real-time WebSocket streaming
CI verification pipeline — Determinism checks and unseeded random scan across all signal operations

📄 License

MIT License — see LICENSE for details.

📞 Support

GitHub Issues | Discussions | PyPI

WiFi DensePose — Privacy-preserving human pose estimation through WiFi signals.

Name		Name	Last commit message	Last commit date
Latest commit History 123 Commits
.claude-flow		.claude-flow
.claude		.claude
.github/workflows		.github/workflows
.swarm		.swarm
assets		assets
docker		docker
docs		docs
firmware/esp32-csi-node		firmware/esp32-csi-node
logging		logging
monitoring		monitoring
plans		plans
references		references
rust-port		rust-port
scripts		scripts
ui		ui
v1		v1
vendor/ruvector		vendor/ruvector
.dockerignore		.dockerignore
.gitignore		.gitignore
.mcp.json		.mcp.json
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
claude.md		claude.md
deploy.sh		deploy.sh
example.env		example.env
install.sh		install.sh
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
verify		verify

What	How	Speed
Pose estimation	CSI subcarrier amplitude/phase → DensePose UV maps	54K fps (Rust)
Breathing detection	Bandpass 0.1-0.5 Hz → FFT peak	6-30 BPM
Heart rate	Bandpass 0.8-2.0 Hz → FFT peak	40-120 BPM
Presence sensing	RSSI variance + motion band power	< 1ms latency
Through-wall	Fresnel zone geometry + multipath modeling	Up to 5m depth

Option	Hardware	Cost	Full CSI	Capabilities
ESP32 Mesh (recommended)	3-6x ESP32-S3 + WiFi router	~$54	Yes	Pose, breathing, heartbeat, motion, presence
Research NIC	Intel 5300 / Atheros AR9580	~$50-100	Yes	Full CSI with 3x3 MIMO
Any WiFi	Windows/Linux laptop	$0	No	RSSI-only: coarse presence and motion

Folders and files

Latest commit

History

Repository files navigation

WiFi DensePose

🚀 Key Features

🏢 Use Cases & Applications

📦 Installation

🚀 Quick Start

1. Install

2. Start the System

3. REST API

4. Real-time WebSocket

📋 Table of Contents

📡 Signal Processing & Sensing

Performance Benchmarks (Validated)

🧠 Models & Training

End-to-End Pipeline

Signal Processing Detail

Deployment Topology

🖥️ CLI Usage

REST API (Rust Sensing Server)

🧪 Testing

🚀 Deployment

Environment Variables

📊 Performance Metrics

Rust Sensing Server

Python vs Rust

🤝 Contributing

📄 Changelog

v2.3.0 — 2026-03-01

v2.2.0 — 2026-02-28

v2.1.0 — 2026-02-28

📄 License

📞 Support

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages