PRAHAR – Real-Time Edge AI Reconnaissance System

YOLOv8 · OpenCV · Flask · React · Three.js

---

🌐 Live Deployment

Shipped and running — no setup required.

Layer	Platform	URL
🖥️ Frontend	Vercel	prahar-eight.vercel.app
⚙️ Backend API	HuggingFace Spaces (Docker)	harsha-tarot117-prahar-backend.hf.space

---

🎯 Overview

PRAHAR processes uploaded drone videos frame-by-frame using YOLOv8, detects military threats (enemy soldiers & vehicles), overlays bounding boxes with confidence scores, and visualizes the results in a military-grade dark themed dashboard with 3D terrain maps.

---

⚙️ Quick Start

Backend (Flask + YOLOv8)

cd prahar/backend
pip install -r requirements.txt
python app.py

Backend starts on http://localhost:5000

Frontend (React + Vite)

cd prahar/frontend
npm install
npm run dev

Frontend starts on http://localhost:5173

---

🧠 Architecture

Upload Video → Flask Backend → OpenCV Frame Extraction
                                    ↓
                             YOLOv8 Inference
                                    ↓
                        Detection Mapping + Bounding Box Draw
                                    ↓
                        Output Video (MP4) + JSON Metrics
                                    ↓
                     React Dashboard ← Axios HTTP Client
                          ↓                  ↓
                   Video Playback       MetricsPanel
                   (with HUD overlay)   TerrainMap (Three.js)

---

🗂️ Project Structure

prahar/
├── backend/
│   ├── app.py              # Flask server, API routes
│   ├── video_processor.py  # YOLOv8 + OpenCV processing pipeline
│   ├── utils.py            # Threat mapping, drawing utilities
│   └── requirements.txt
├── frontend/
│   └── src/
│       ├── App.jsx                    # Root component & layout
│       ├── index.css                  # Military dark theme CSS
│       └── components/
│           ├── MetricsPanel.jsx       # Animated stats + threat level
│           ├── TerrainMap.jsx         # Three.js 3D terrain
│           └── ActivityLog.jsx        # Live event log
└── README.md

---

🔍 Detection Mapping

YOLO Class	Threat Type	Marker Color
person	Enemy Soldier	🔴 Red
car, truck	Military Vehicle	🟡 Yellow
bus, boat	Military Vehicle	🟡 Yellow
airplane	Military Vehicle	🟡 Yellow

---

⚠️ Threat Level Logic

Condition	Threat Level	UI Color
Any enemy soldiers detected	HIGH	🔴 Red flashing
Only military vehicles	MEDIUM	🟡 Yellow
No threats	LOW	🟢 Green

---

📊 API Reference

POST /analyze-video

Upload a drone video for threat analysis.

Request: multipart/form-data with field video

Response:

{
  "video_url": "/video/output_abc123.mp4",
  "threat_level": "HIGH | MEDIUM | LOW",
  "total_detections": 42,
  "enemy_count": 12,
  "vehicle_count": 8,
  "fps": 18.4,
  "duration": 34.5,
  "frame_count": 863,
  "resolution": "1920x1080",
  "threat_coordinates": [
    { "x": 15.5, "y": -20.1, "z": 2.1, "type": "Enemy Soldier", "confidence": 87.3 }
  ]
}

GET /video/<filename>

Serve the processed output video.

GET /health

System health check.

---

🌍 3D Terrain Visualization

Built with Three.js — generates a procedural terrain using sinusoidal noise:

• Green terrain → elevation-colored geometry
• Wireframe overlay → military grid aesthetic
• Auto-rotating camera → continuous orbit around the battlefield
• Red markers → Enemy Soldier positions
• Yellow markers → Military Vehicle positions
• Pulsing rings → animated threat indicators

---

🔗 Datature Integration (Conceptual)

This section explains how PRAHAR can scale from a prototype to a production-ready military AI system using Datature's annotation and training platform.

What is Datature?

Datature is a no-code AI platform for building, training, and deploying computer vision models. It provides:

• Annotation Tools — Bounding box, polygon, semantic & instance segmentation
• Model Training — Train YOLOv8, Detectron2, EfficientDet on custom datasets
• Model Hub — Version-controlled model registry
• Deployment — One-click inference API deployment

How Datature Fits Into PRAHAR

Stage	Tool	Description
Dataset Curation	Datature Annotator	Label drone footage with enemy/vehicle classes
Model Training	Datature Nexus	Fine-tune YOLOv8 on military terrain images
Model Export	Datature Hub	Export .pt model → deploy to edge device
Edge Inference	PRAHAR Backend	Run Datature model via Ultralytics API

Replacing YOLOv8 Generic with Datature-Trained Model

# Current: Generic YOLOv8 pretrained on COCO
model = YOLO("yolov8n.pt")

# Scalable: Datature-trained military detection model
model = YOLO("datature_military_v2.pt")  # Exported from Datature Hub

The swap is one line — the entire PRAHAR pipeline scales automatically.

Benefits

• Terrain-specific accuracy — Model trained on camouflage, night-vision, aerial footage
• Custom class vocabulary — sniper, armored_vehicle, bunker, weapon_cache
• Continual learning — Retrain via Datature as new footage is labeled
• Edge-optimized — Export TFLite/ONNX for Jetson/Raspberry Pi deployment

---

🎨 UI Features

• Military dark theme with green accent colors (#00ff8c)
• Real-time clock in UTC
• Blinking HIGH threat alert on video frame
• Animated metric counters (count-up animation)
• Drag & drop video upload
• HUD overlay on processed video frames (brackets, crosshair, progress bar, timestamp)
• Activity log with timestamped events
• 3D rotating terrain with threat markers

---

🚀 Demo Mode

If ultralytics is not installed or the model fails to load, PRAHAR automatically enters Demo Mode:

• Generates synthetic detections per frame
• Still draw bounding boxes, HUD, metrics
• Full UI experience without real ML inference
• Great for showcase without a GPU

---

🖥️ System Requirements

Component	Minimum	Recommended
Python	3.10+	3.11+
RAM	8 GB	16 GB
GPU	None (CPU mode)	NVIDIA 8GB VRAM
Storage	2 GB	10 GB
Node.js	18+	20+

---

📄 License

MIT License — Free for hackathon use.

---

Built for the PRAHAR Defense-Tech Hackathon · Edge AI · Drone Surveillance · Real-Time Threat Detection