A capstone project focused on creating an AI-powered video synopsis system that condenses long security camera footage into short, informative highlight videos using computer vision and deep learning.
Security cameras generate hours of video footage daily, making manual review time-consuming and inefficient. Our system solves this by implementing a video synopsis technique that extracts and compiles only the significant events—such as human movement—into a concise summary video.
The project is built using a modular architecture involving:
- Frontend: React.js
- Backend: Node.js + Express
- Database: MongoDB
- AI Module: YOLOv8, OpenCV, PyTorch
The AI module was developed using:
- YOLOv8 for real-time object detection
- OpenCV for image and video processing (frame extraction, motion analysis, etc.)
- PyTorch for model training and fine-tuning
- FFmpeg for video encoding and optimization
The system identifies and tracks moving objects (primarily humans), and merges key moments into a summarized clip, preserving spatial-temporal integrity.
- 🎯 Real-time object detection and tracking
- 📦 Video summarization using frame-based motion analysis
- ⚡ Fast video processing with optimized pipelines
- 🔐 Privacy-focused and scalable
- 💻 User-friendly web interface for uploading and viewing videos
- Automate video review for surveillance footage
- Reduce review time and storage needs
- Enable faster identification of critical moments in long videos
- Build a deployable, user-friendly platform for end users
| Layer | Technologies |
|---|---|
| Frontend | React.js |
| Backend | Node.js, Express |
| Database | MongoDB |
| AI Module | YOLOv8, PyTorch, OpenCV, FFmpeg |
- Hasan Kemal Mete (Computer Vision & AI)
- Bengühan Şahin (Computer Vision & AI)
- Ekrem Bulut (Frontend & Backend Development)
- Doğa Yıldız (Software Engineering)
- Sarper Sarp (Software Architecture & Integration)