Skip to content
takenrohitPublic

About

Real-time distress detection from live CCTV feeds · YOLOv8 · MediaPipe · Gemini Vision · FastAPI

Resources

Stars

3 stars

Watchers

0 watching

Forks

Latest commit

 

History

86 Commits

Folders and files

Repository files navigation

🛡️ Argus — Real-Time Public Safety Analytics Argus is an AI-powered real-time public safety surveillance system designed to detect distress situations from CCTV feeds and automatically alert authorities.

Built during a 24-hour national-level hackathon.

🚨 Problem Statement Existing CCTV systems are passive — they record incidents but do not respond in real time.

In critical situations like stalking, harassment, or assault, every second matters.

Argus transforms passive CCTV infrastructure into an intelligent, real-time threat detection system focused on women’s safety in public spaces.

🎯 Our Solution Argus connects to live camera feeds and uses AI-based spatial and motion analysis to detect high-risk distress scenarios.

When suspicious activity is detected, the system:

✅ Classifies threat level ✅ Generates real-time alert ✅ Sends notification to dashboard ✅ Logs incident with timestamp and confidence

🧠 Core Features 1️⃣ Encirclement Detection Detects when:

One person is surrounded by 3 or more individuals Movement becomes restricted Proximity threshold maintained across frames Use case: Group intimidation / harassment scenario.

2️⃣ Following / Pursuit Detection Detects when:

One individual consistently follows another Maintains close distance Mirrors directional changes Use case: Stalking detection.

3️⃣ Physical Struggle Detection Detects when:

Two individuals are in very close proximity Bounding boxes overlap significantly Rapid erratic movement occurs Use case: Assault detection.

🏗️ System Architecture text

Video Feed ↓ YOLOv8 Person Detection & Tracking ↓ Distress Detection Engine ↓ Alert Manager ↓ WebSocket Broadcast ↓ Live Dashboard

⚙️ Tech Stack

AI & Computer Vision Ultralytics YOLOv8 (person detection & tracking) OpenCV (video processing) Backend FastAPI WebSockets Frontend Next.js TailwindCSS Leaflet.js (map view) Optional AI Verification Gemini Vision API (scene validation)

📁 Project Structure

Argus/ │ ├── backend/ │ ├── main.py │ ├── core/ │ │ ├── yolo_tracker.py │ │ ├── distress_engine.py │ │ └── video_processor.py │ ├── alerts/ │ └── api/ │ ├── frontend/ │ ├── app/ │ ├── components/ │ ├── assets/ │ └── demo_videos/ │ └── README.md

🚀 How to Run

Backend

1️⃣ Install backend dependencies

pip install -r requirements.txt

2️⃣ Start FastAPI from the repo root

uvicorn backend.main:app --reload

Frontend

1️⃣ Install frontend dependencies

cd frontend/Argus2.0/ASSETS/argus-smart-surveillance && npm install

2️⃣ Start the Vite app during development

npm run dev

The Vite dev server now proxies /api and /ws to the FastAPI app on http://127.0.0.1:8000.

Single-app build

1️⃣ Build the frontend

cd frontend/Argus2.0/ASSETS/argus-smart-surveillance && npm run build

2️⃣ Start FastAPI again

uvicorn backend.main:app --reload

FastAPI will automatically serve the built frontend from the generated dist/ folder.

🛡️ Supabase Setup (Important)

To ensure the dashboard works correctly, please configure your Supabase project:

  1. Tables: Create an incidents table (see backend/models/incident.py for schema).
  2. RLS Policies: In the Supabase dashboard, set the incidents table policy to "Enable access for all users" for SELECT, INSERT, and UPDATE (since we use the anon key).
  3. Storage: Create a public bucket named incidents for evidence screenshots. 🎥 Running Video Detection The system supports:

✅ Webcam ✅ IP Camera (RTSP) ✅ Pre-recorded video

Example:

Python

tracker = YOLOTracker("backend/sample_videos/incident.mp4") tracker.run() 📊 Dashboard Pages / → Live Camera Feed + Active Alerts /incidents → Incident History /map → Camera Location Map 🔒 Privacy & Design Philosophy No facial recognition used No identity storage Focus on behavioral and spatial risk detection Designed to integrate with existing CCTV infrastructure 🧩 Why Argus? ✅ Real-time detection ✅ No new hardware required ✅ AI-assisted decision support ✅ Scalable to smart city infrastructure ✅ Designed specifically for women safety escalation scenarios

📈 Future Improvements

Multi-camera distributed processing Crowd behavior modeling Facial distress signal modeling (high-resolution feeds) Integration with emergency dispatch systems Mobile officer app 🤝 Open-Source & AI Disclosure This project uses:

Ultralytics YOLOv8 (pretrained model)

OpenCV FastAPI Next.js AI coding assistance tools during development All integration and system logic were implemented during the hackathon period.

👥 Team

Built by Team Code Cruzers

Team members: Shresth Kumar Rohit Shukla Syed Hussain

🛡️ Tagline “From Passive Surveillance to Proactive Protection.”

About

Real-time distress detection from live CCTV feeds · YOLOv8 · MediaPipe · Gemini Vision · FastAPI

Resources

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages