🛡️ Kavach - Vandalism Detection Web Application

A modern, beautiful web interface for vandalism detection with person and pose estimation using AI.

Features

• 📸 Image Upload & Processing - Upload images and detect vandalism, persons, and poses
• 🎥 Video Upload & Processing - Process videos frame by frame
• 🎨 Modern UI/UX - Beautiful, responsive design with dark theme
• 📊 Confidence Threshold Control - Adjustable detection sensitivity (1-100%)
• 📁 Past Runs Gallery - View all previously processed results
• ⚡ Real-time Processing - Fast inference with YOLOv11 and MediaPipe

Installation

1. Activate virtual environment:

source env/bin/activate

2. Install dependencies:

pip install -r requirements.txt

3. Ensure models are in place:
   • Vandalism model: train/weights/best.pt
   • Human detection model: yolo11n.pt

Running the Application

1. Start the Flask server:

python app.py

2. Open your browser: Navigate to http://localhost:5000

Usage

1. Upload Media:

   • Click the upload area or drag & drop an image/video file
   • Supported formats: PNG, JPG, JPEG, GIF, BMP, MP4, AVI, MOV, MKV

2. Adjust Settings:

   • Use the confidence slider to adjust detection sensitivity
   • Lower values = more detections (may include false positives)
   • Higher values = fewer detections (more strict)

3. Process:

   • Click "Process" to run inference
   • Results will appear below

4. View Past Runs:

   • Scroll down to see gallery of all previous results
   • Click any item to view full size

API Endpoints

• GET / - Main web interface
• POST /api/upload - Upload and process file
• GET /api/past-runs - Get list of past results
• GET /api/results/<filename> - Download result file
• GET /api/health - Health check

File Structure

vandalism/
├── app.py                 # Flask application
├── templates/
│   └── index.html        # Main HTML template
├── static/
│   ├── css/
│   │   └── style.css     # Stylesheet
│   └── js/
│       └── app.js        # JavaScript
├── uploads/              # Temporary upload storage
└── runs/detect/inference/ # Results storage

Output Examples

Example 1: Vandalism Detection on Brick Wall

Graffiti detected on a brick wall with confidence scores displayed.

Example 2: Multiple Graffiti Detections

Detection of multiple graffiti elements with individual bounding boxes and confidence scores

Example 3: Combined Detection (Vandalism + Person + Pose)

Complete detection showing vandalism (red boxes), person detection (red boxes), face detection (yellow boxes), and pose estimation (skeleton overlay)

Notes

• Models are loaded lazily on first request (may take a moment)
• Uploaded files are automatically cleaned up after processing
• Results are saved in runs/detect/inference/ directory
• Maximum file size: 500MB

Troubleshooting

• Models not loading: Ensure train/weights/best.pt and yolo11n.pt exist
• Port already in use: Change port in app.py (last line)
• Memory issues: Reduce video resolution or process shorter clips