360° Spherical Panorama Stitching

Transform your phone photos into stunning 360° immersive panoramas with professional-grade quality. This project automatically stitches together a sequence of images captured while rotating your phone, creating seamless equirectangular panoramas that can be viewed in VR headsets, web browsers, or panoramic viewers.

🎬 Showcase Video:

✨ Features

• 🎯 Automatic Processing: Simply provide a sequence of photos or a video, and the pipeline handles everything
• 📱 Phone Camera Optimized: Works with any phone camera, automatically extracts camera parameters from EXIF data
• 🎬 Video Support: Extract frames from videos automatically with intelligent keyframe selection
• 🔍 Robust Matching: Advanced feature detection and matching handles challenging scenes with low contrast or repetitive patterns
• 🎨 Multiple Blending Modes: Choose from sharp blending (minimal blur) or smooth blending (seamless transitions)
• 🌐 Interactive Viewer: Automatically generates a web-based 360° viewer using Three.js
• ⚡ Memory Efficient: Handles large panoramas (8K+) with optimized processing
• 🛠️ Fully Configurable: Fine-tune every aspect through a simple YAML configuration file

🎨 What Makes This Special?

Unlike simple panorama apps, this tool:

• Uses computer vision algorithms (ORB features, RANSAC, homography estimation) for robust alignment
• Extracts pure rotation from homographies, ensuring geometrically correct panoramas
• Handles failed matches intelligently with neighbor interpolation and temporal smoothing
• Supports high-resolution output (up to 8K+ for professional use)
• Optimized for video input with frame extraction and motion-based keyframe selection

🚀 Quick Start

# Install dependencies
pip install -r requirements.txt

# Create a config file (see USAGE.md for details)
# Edit config.yaml with your settings

# Run the pipeline
python run.py config.yaml

# View your panorama
open output/viewer/index.html

For detailed usage instructions, see USAGE.md.

📊 Technical Highlights

This implementation features:

• Pure rotation model - Assumes camera rotates around its optical center (no parallax)
• Spherical projection - Maps images to equirectangular format using inverse mapping
• Temporal smoothing - Smooths camera rotation estimates for video sequences
• Neighbor interpolation - Handles failed feature matches intelligently
• Adaptive RANSAC - Adjusts threshold based on image resolution
• Memory optimization - Sequential processing for large inputs

For a deep technical dive, see TECHNICAL.md.

📸 Sample Output

The generated panoramas can be:

• Viewed in browsers using the included HTML viewer
• Uploaded to platforms like YouTube (360°), Facebook, or Instagram
• Imported into VR applications for immersive experiences
• Used in professional workflows for architectural visualization or virtual tours

🎯 Perfect For

• Photographers creating immersive experiences
• Real estate virtual tours
• Travel content creators
• Architectural visualization
• VR/AR developers needing 360° content
• Educators teaching computer vision concepts

🔧 Requirements

• Python 3.8+
• OpenCV 4.8+
• NumPy, Pillow, and other dependencies (see requirements.txt)

📚 Documentation

• USAGE.md - Complete usage guide and configuration reference
• TECHNICAL.md - Deep technical dive into algorithms and implementation

🤝 Contributing

Contributions are welcome! This is an open-source project built with passion for computer vision and panoramic imaging.

📄 License

MIT License - see LICENSE for details.

🙏 Acknowledgments

• OpenCV for computer vision algorithms
• Three.js for the 360° viewer framework
• Inspired by classic panorama stitching research and implementations

---

Made with ❤️ for the panoramic imaging community