Telemetria Sensori
Cam Trap
12
Nodi Audio
8
Collari IoT
34
Agg. OSINT
1.4K
Allarmi Minaccia Attivi
CAM_TRAP_04 15:42 UTC
Rilevamento Umano Armato (Bracconiere)
CONF: 94%
AUDIO_NODE_02 15:39 UTC
Rilevamento Picco Energetico (Sparo)
CONF: 78%
An open-source anti-poaching command center —
built on free tools, runs offline, costs $0 in licenses.
10 capability modules 20+ repos integrated works without internet no vendor lock-in
📷
Detect poachers on camera traps

AI runs on-device. Alert in 30s when a human appears. No cloud, no subscription.

M2 Playbook →
🔊
Hear gunshots 3 km away, at night

Cheap microphone network. TDoA triangulation sends GPS coordinates in <30s.

M3 Playbook →
🕵️
Monitor illegal wildlife markets

Automated scrapers watch Telegram and classifieds. Price spikes predict seizures.

M7 Playbook →
⚖️
Evidence that holds up in court

SHA-256 hash chain on every file. Auto-generated chargesheet with CITES sections.

M9 Playbook →
🗺️
Patrol routes poachers can't predict

PAWS + RL generates a new route every day from historical incident data.

M8 Playbook →
🔓
Fully open, fully yours

Fork it, adapt it, deploy it. No accounts, no API keys, no paywalls.

GitHub →
Each module solves one field problem independently — deploy one, all, or any combination. The Essential tier of every module works with a phone and a free app. Full playbooks with code, hardware lists, and agent prompts are in docs/.

📚 Operator Playbooks & Toolkit — the real content lives in docs/

This dashboard is an illustrative demo. The actionable material — one playbook per capability module (recommended repos, runnable Python training/inference scripts, Arduino node firmware, agent prompts, milestones, international use cases) — is in the repository:

  • docs/README.md — master index + recommended-repo catalog + use-case matrix
  • docs/modules/M01..M10.md — 10 capability playbooks (M2 vision & M3 audio fully fleshed)
  • toolkit/python/ — camera-trap & audio trainers, edge inference, TDoA localizer
  • toolkit/arduino/ — PIR camera node + acoustic node firmware
  • docs/prompts/ — paste-into-any-agent prompt packs per milestone

Clone the hub → pick your use case → follow the playbook with free tools. Wildlife can't wait.

🧭 What you can build — 10 capability modules

Each module is a complete playbook: what to build, which open repos to use, runnable code, field guides. Pick the one that fits your situation and follow the steps.

M1 · Set up the platform

Get the hub running, define user roles, connect it to existing tools your team already uses (SMART, EarthRanger, local DB).

→ playbook
M2 · See what cameras see — automatically

Camera traps detect humans, weapons and vehicles on-device — no cloud needed. Night vision and false-positive filtering included. Photos auto-tagged as evidence.

→ full playbook + code
M3 · Hear gunshots and chainsaws 24/7

Cheap microphone nodes detect gunshots, chainsaws and vehicles in the dark. Multiple nodes pinpoint the location. Works offline.

→ full playbook + code
M4 · Drone and satellite overwatch

Schedule drone patrols, add thermal night flights, layer satellite maps to spot habitat changes and escape routes before rangers deploy.

→ playbook
M5 · Track animals, not just poachers

GPS collars send alerts when animals leave safe zones. Detect when a herd stops moving — often the first sign of a poaching event.

→ playbook
M6 · Patrols and community tips

Rangers log incidents offline with no phone signal. Local community members report sightings anonymously. All data syncs when connectivity returns.

→ playbook
M7 · Monitor illegal wildlife markets

Watch online marketplaces and social media for wildlife trade. Geolocate photos. Map trafficking networks. Get early warning when prices spike for a protected species.

→ playbook
M8 · Predict where poachers strike next

Historical incident data generates risk maps. AI optimises patrol routes so rangers cover high-risk areas first. Updated daily with moon phase and weather.

→ playbook
M9 · Build court-ready evidence

Every alert generates a tamper-proof evidence file with timestamp and location. Auto-generate incident reports compatible with CITES and national wildlife law. See live demo above.

→ playbook
M10 · Keep it running long-term

Monitor sensor battery life, catch failures before they happen. Train new rangers with scenario exercises. Keep AI models improving without sharing raw data.

→ playbook

Full repo catalog & use-case matrix: docs/README.md

Hardware Node Reference — Arduino TinyML (M2 / M3)

Low-cost camouflaged sensor nodes built on Arduino Nicla Vision or ESP32-CAM. A PIR sensor wakes the microcontroller from deep sleep (<50 µA), captures a frame, runs on-device TinyML inference, and sends a compressed LoRa packet only if confidence exceeds 80%.

Hardware Wiring:
- PIR sensor VCC → 3.3V / Out → GPIO 13 (Wakeup)
- GC2145 microcamera → dedicated DVP interface
- LoRa module → GPIO 10 (SPI SS), 11 (MOSI), 12 (MISO), 13 (SCK)
- Solar cell + LiPo TP4056 charger on-board
Physical camouflage (anti-tagging):
3D-printed PLA biodegradable shell coated with a hyper-realistic bark-texture mix matched to the target forest. AI-generated textures per deployment region. Transmitters shut off immediately after send to prevent scanner detection.
Reference firmware (Nicla Vision / ESP32):
#include <Nicla_System.h> #include <EloquentTinyML.h> // TensorFlow Lite edge wrapper #include "model_yolo_quantized.h" // quantised TinyML model #define PIR_PIN 13 #define LORA_SS 10 void setup() { nicla::begin(); nicla::leds.begin(); pinMode(PIR_PIN, INPUT_PULLDOWN); lowPowerConfigureWakeup(PIR_PIN, RISING); // PIR wakes board } void loop() { nicla::deepSleep(); // <50 µA until PIR fires Camera.captureFrame(); float predictions[3]; // Human | Animal | Background ml.predict(Camera.frameBuffer, predictions); if (predictions[0] > 0.80) { // human confidence >80% LoRa.sendTacticalEvent("CAM_EDGE_42", "poacher", predictions[0]); } }

Production Readiness Baseline

The project must remain useful before every subsystem is complete. The first production target is a dependable command-center shell with local persistence, clear fallback behavior, and a documented operational path.

Real-World Operational Playbooks

These guides describe how the platform should be used in concrete field scenarios. They are written for rangers, operators, and maintainers, not for architecture diagrams.

Documentation Scaling Plan

dev.md is the master technical map. As recurring workflows stabilize, each use case should move into its own focused document so the project can be extended without losing clarity.

WildGuard integrates existing best-in-class open-source tools rather than rebuilding from scratch. Each cluster below maps repos to the module they power. Full integration notes and code references live in docs/.

Computer Vision & Detection — M2

microsoft/CameraTraps MegaDetector v5
State-of-the-art human/animal/vehicle detector. Base model for M2 — triggers an alert when a human appears at >80% confidence.
microsoft/pytorch-wildlife
Wildlife classifier hub. Used to load and fine-tune species-specific detectors for M2 and M5 without rewriting the training pipeline.
conservationxlabs/animl-frontend
ML-assisted camera trap review UI. UI pattern adopted for M2's triage interface — rangers review only high-confidence flagged images.

Bioacoustics — M3

rfcx/arbimon production
Rainforest Connection's audio monitoring platform. Architecture reference for M3's microphone node format and audio stream ingestion schema.
bioacoustics-open-source/opensoundscape
Python audio ML toolkit. Powers M3's spectrogram pipeline and Keras model runner for gunshot/chainsaw classification in audio_dsp.py.

Aerial & Geospatial — M4

OpenDroneMap/WebODM
Drone photogrammetry to ortophoto pipeline. Generates updated terrain maps for M4 patrol commanders before each mission.
GBIF · IUCN Red List · Copernicus Sentinel-1 · Open-Meteo
Free data APIs used as enrichment layers across M4, M5, M7, M8: species range, habitat type, SAR imagery, hourly weather.

Species Intelligence — M5

WildMeOrg/Wildbook
Photo-ID platform for individual animal recognition without physical collars. M5 uses it to build population models from camera trap images.
MammAlps / MEWC (zaandahl)
Alpine mammals dataset and classification benchmark. Used to validate M2 and M5 models on European mountain species (ibex, wolf, bear).

Prediction & Patrol Routing — M8

lily-x/PAWS-public deployed Uganda
Game-theoretic anti-poaching patrol scheduler. M8 wraps PAWS's iware/ + planning/ modules as a FastAPI microservice generating daily routes from historical incident data.
iamlab-cmu/wildlifeRL
DDPG reinforcement learning for patrol unpredictability. Combined with PAWS in M8 so rangers don't follow routes that poachers have memorised.

Evidence & Legal Chain — M9

SHA-256 Hash Chain (built-in)
Every media file hashed at capture time. Chain stored locally. M9 auto-generates chargesheet PDFs citing the correct CITES schedule articles for the seized species.
audtheia/audtheia-environmental-monitoring
Environmental event auditing schemas. Provides the evidence format and field report structure M9 uses to reconstruct incident timelines for prosecution.

Proposta Istituzionale per WWF, Riserve & Guardie Forestali

WildGuard AI è concepito per essere presentato a ONG ed enti governativi come soluzione sostenibile a basso costo. Fornisce un ecosistema che garantisce la catena di custodia legale delle prove, ottimizza il coordinamento con logica di gioco stocastico e riduce del 90% i falsi allarmi sul campo.

1. Sostenibilità Economica e Zero Licenze
Utilizza esclusivamente software open-source e hardware commerciale standard (Arduino/ESP32). Nessun canone di abbonamento o vincolo con vendor di tecnologie di sorveglianza proprietarie.
2. Validità delle Prove in Tribunale
L'evidenza multimediale viene registrata con un hash crittografico SHA-256 e legata ad una catena locale immutabile (hash chain). Questo assicura l'integrità e previene obiezioni legali sulle foto delle camera trap durante i procedimenti penali.
3. Sottile Sicurezza sul Campo e Anti-Tracciamento
I nodi periferici spengono i moduli trasmittenti subito dopo l'invio (anti-tagging). Questo impedisce ai bracconieri dotati di scanner di localizzare le posizioni dei ranger o le postazioni di guardia.
Stato dei Sotto-Repository
FaunaScope (TrapSense) ONLINE
Forest-Conservation (VanaRakshya) ONLINE
PAWS-public (Predictor) ONLINE
Wildlife-News (OSINT Parser) ONLINE
WildlifeFL (Federated Learning) DEGRADED
Intelligence Feed