Skip to content
BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
app
app
 
 
backend-disaster
backend-disaster
 
 
client
client
 
 
gundetection
gundetection
 
 
landing
landing
 
 
lost-and-found
lost-and-found
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 
fix_secrets.sh
fix_secrets.sh
 
 

Repository files navigation


🔱 Trinetra

AI-Powered Public Safety Platform for Large-Scale Events

Real-time crowd intelligence · Weapon detection · Disaster prediction · Missing person tracing · Emergency response

Overview

Trinetra ("three eyes" in Sanskrit) is a full-stack, AI-first public safety platform purpose-built for high-density events — pilgrimages, melas, and mass gatherings like Pandharpur Wari and Mahakumbh. It gives authorities a single pane of glass to monitor crowds, detect threats, predict disasters, and coordinate emergency response in real time.

The system is split across five independent services that work together:

Service What it does
app/ Expo React Native mobile app for citizens and field admins
client/ Vite + React web dashboard for command-centre operators
gundetection/ Flask microservice — weapon & threat detection via Gemini Vision
backend-disaster/ Flask microservice — disaster prediction, weather, earthquake, satellite
lost-and-found/ Flask microservice — YOLO-based missing person detection
landing/ Next.js public marketing site

Architecture

┌─────────────────────────────────────────────────────────────┐
│                         Clients                             │
│  ┌──────────────────┐          ┌──────────────────────────┐ │
│  │  Expo Mobile App │          │  Vite React Web Dashboard│ │
│  │  (citizens +     │          │  (command centre admins) │ │
│  │   field admins)  │          └──────────────────────────┘ │
│  └──────────────────┘                                        │
└──────────────────────────┬──────────────────────────────────┘
                           │ Firebase Auth + Firestore
            ┌──────────────┼──────────────┐
            │              │              │
   ┌────────▼─────┐ ┌──────▼──────┐ ┌───▼──────────────────┐
   │ Crowd Agent  │ │ Disaster    │ │ Weapon Detection     │
   │ (Agno/Gemini)│ │ Prediction  │ │ (Gemini Vision)      │
   │ Port 5000    │ │ Port 5001   │ │ Port 5002            │
   └──────────────┘ └─────────────┘ └──────────────────────┘
                           │
                    ┌──────▼──────┐
                    │ Lost & Found│
                    │ (YOLO)      │
                    │ Port 5003   │
                    └─────────────┘

Features

🧠 AI Detection Engine

  • Weapon & Threat Detection — Gemini 2.5 Flash analyses CCTV frames to classify guns, knives, fire, fighting, and suspicious behaviour. Returns per-frame risk level (safe → anomaly → danger → critical) with screenshot evidence.
  • Crowd Density Analysis — Agno LLM agent powered by Gemini 1.5 Flash processes video frames and returns structured crowd assessments: headcount estimate, stampede risk, police/medical staffing recommendations, and chokepoint detection.
  • Face Recognition / Missing Persons — YOLO-v8 person detector combined with face embeddings to match uploaded reference photos against CCTV footage. Powers both the app's missing person reporting and the admin search portal.

🗺️ Live Geofenced Map

  • Real-time location tracking for citizens attending events
  • Geofence zones with automatic breach alerts sent to family contacts
  • Evacuation route overlays, assembly points, and facility finder
  • Google Maps integration on iOS/Android with offline fallback

🌩️ Disaster Prediction

  • Weather — OpenWeatherMap current conditions + 5-day forecast with configurable alert thresholds
  • Earthquakes — USGS public feed with magnitude filtering and proximity-to-event checks
  • Satellite Imagery — Sentinel Hub flood analysis and terrain mapping via bounding box queries
  • Risk Score Engine — Composite risk calculation across weather, seismic, crowd, and traffic signals, broadcast over WebSocket to all connected dashboards
  • SOS Reporting — Pilgrim distress reports geotagged and surfaced to admin in real time

🏥 Medical & Emergency Response

  • Ambulance dispatch request flow with live status tracking
  • Nearby hospital lookup with geo-radius query
  • Doctor video queue for remote triage (VideoSDK integration)
  • Emergency offline mode — core SOS works without internet

🔍 Lost & Found

  • Citizens upload a photo of a missing person; YOLO detects and crops all persons from crowd footage
  • Face similarity matching returns top candidates with confidence scores
  • Admin portal to manage reports, update status, and reunite families

Repository Structure

trinetra/
├── app/                          # Expo React Native mobile app
│   ├── app/                      # Expo Router screens
│   │   ├── (tabs)/               # Citizen-facing tab screens
│   │   ├── (adminTabs)/          # Field admin tab screens
│   │   ├── (medicalAdminTabs)/   # Medical admin screens
│   │   ├── LoginScreen.tsx
│   │   └── RegisterScreen.tsx
│   ├── lib/                      # Firebase, hospital, notification services
│   ├── components/               # Shared UI primitives
│   ├── app.py                    # Crowd analysis LLM agent (Flask)
│   └── app.json                  # Expo app manifest
│
├── client/                       # Vite + React web dashboard
│   └── src/
│       ├── pages/                # All dashboard pages
│       │   ├── disaster/         # Disaster module sub-pages
│       │   ├── medical/          # Medical module sub-pages
│       │   └── doctor/           # Doctor portal sub-pages
│       ├── components/           # Shared layout, UI, animated components
│       └── context/              # AuthContext, DisasterDataContext
│
├── gundetection/                 # Weapon/threat detection microservice
│   ├── main.py                   # Flask API + Gemini Vision pipeline
│   ├── requirements.txt
│   └── render.yaml
│
├── backend-disaster/             # Disaster prediction microservice
│   ├── app.py                    # Flask REST API + WebSocket
│   ├── config.py                 # City configs, API keys, thresholds
│   └── services/
│       ├── weather_service.py
│       ├── earthquake_service.py
│       ├── satellite_service.py
│       ├── crowd_service.py
│       ├── traffic_service.py
│       ├── alert_service.py
│       ├── pilgrim_service.py
│       └── data_store.py         # SQLite persistence layer
│
├── lost-and-found/               # YOLO person detection microservice
│   ├── app.py                    # Flask API
│   ├── person_detector.py        # YOLOv8 + face matching logic
│   └── Dockerfile
│
└── landing/                      # Next.js public landing page

Getting Started

Prerequisites

  • Node.js ≥ 18
  • Python ≥ 3.10
  • Expo CLI (npm install -g expo-cli)
  • Firebase project with Auth + Firestore enabled

1. Mobile App (app/)

cd app
npm install
npx expo start

Scan the QR code with Expo Go on iOS or Android, or press a for Android emulator / i for iOS simulator.

Key environment variables — set inside app.json or via EAS:

GOOGLE_MAPS_API_KEY=...
FIREBASE_API_KEY=...

2. Web Dashboard (client/)

cd client
npm install
npm run dev           # starts at http://localhost:5173

3. Weapon Detection Service (gundetection/)

cd gundetection
pip install -r requirements.txt
cp env.example .env   # fill in GEMINI_API_KEY
python main.py        # runs on port 5002

Endpoints:

  • POST /analyze — upload image or video; returns per-frame threat analysis
  • GET /health — service health check

4. Disaster Prediction Service (backend-disaster/)

cd backend-disaster
pip install -r requirements.txt
cp env.example .env   # fill in OPENWEATHER_API_KEY, GOOGLE_MAPS_API_KEY, SENTINEL_HUB_CLIENT_ID/SECRET
python app.py         # runs on port 5001

Key endpoints:

GET  /api/dashboard          — full dashboard snapshot
GET  /api/weather            — current weather
GET  /api/earthquakes        — USGS seismic data
GET  /api/crowd/stampede-risk
GET  /api/satellite/flood-analysis
POST /api/pilgrim/sos        — submit SOS report
WS   /socket.io              — real-time dashboard updates

5. Lost & Found Service (lost-and-found/)

cd lost-and-found
pip install -r requirements.txt
python app.py         # runs on port 5003

Or with Docker:

docker build -t trinetra-laf .
docker run -p 5003:5003 trinetra-laf

6. Landing Page (landing/)

cd landing
npm install
npm run dev           # starts at http://localhost:3000

Deployment

All Python microservices include a render.yaml for one-click deployment on Render. The lost-and-found service also ships with a Dockerfile and Procfile for Heroku/Railway.

Service Deployed URL
Weapon Detection https://gun-detection.onrender.com
Disaster Prediction configure via REACT_APP_DISASTER_API
Lost & Found configure via REACT_APP_LAF_API

Tech Stack

Layer Technology
Mobile Expo (React Native), Expo Router, TypeScript
Web Dashboard Vite, React, TypeScript, Tailwind CSS, shadcn/ui
Landing Page Next.js 15, TypeScript
AI / LLM Google Gemini 2.5 Flash (vision), Agno agent framework
Computer Vision OpenCV, YOLOv8 (Ultralytics)
Backend Flask, Flask-SocketIO, Flask-CORS, Gunicorn
Database Firebase Firestore (app data), SQLite (disaster events)
Auth Firebase Authentication
Maps Google Maps (mobile + web)
External APIs OpenWeatherMap, USGS Earthquake, Sentinel Hub, Cloudinary
Push Notifications Expo Push Notification Service
Deployment Render, Docker, Railway

Environment Variables Reference

gundetection/.env

GEMINI_API_KEY=your_key
PORT=5002

backend-disaster/.env

OPENWEATHER_API_KEY=your_key
GOOGLE_MAPS_API_KEY=your_key
SENTINEL_HUB_CLIENT_ID=your_id
SENTINEL_HUB_CLIENT_SECRET=your_secret
SECRET_KEY=your_flask_secret
DEFAULT_CITY=pandharpur

Contributing

  1. Fork the repo
  2. Create a feature branch (git checkout -b feat/your-feature)
  3. Commit your changes following Conventional Commits
  4. Open a pull request

License

MIT © 2025 Samarth Kolarkar

Built for MahaKumbh · Pandharpur Wari · and every gathering where safety can't wait.

About

Real-time AI safety platform for mass gatherings — weapon detection, missing person tracing, disaster prediction, and emergency response for events like Mahakumbh and Pandharpur Wari

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages