Video Streaming
- RTMP ingestion from OBS Studio
- FFmpeg HLS conversion
- Event-driven segment detection
- WebSocket-powered live updates
System Architecture
View Full
Complete system overview showing video ingestion, AI processing pipeline, and real-time dashboard components.
Components
- Video Ingestion: RTMP streams from OBS, FFmpeg HLS conversion
- AI Processing: CLIP/YOLO for logo detection and scene analysis
- Real-time Communication: Django Channels WebSocket for live updates
- Frontend: Angular 17+ with HLS.js video player and Canvas overlays
Key Features
AI Analysis
- Logo/brand detection (CLIP)
- Object detection (YOLO)
- Real-time vs batch processing modes
- Switchable local/cloud backends
Infrastructure
- Docker containerized services
- Kubernetes orchestration
- GCP integration (Storage, Vision)
- Celery task queue with Redis
Technology Stack
Backend
- Django + Channels
- Django REST Framework
- PostgreSQL
- Celery + Redis
- FFmpeg
AI/ML
- OpenCV
- CLIP (scene analysis)
- YOLO (object detection)
- Hugging Face Transformers
- GCP Vision API
Frontend
- Angular 17+
- HLS.js video player
- Canvas overlays
- WebSocket client
- Standalone components
Architecture Goals
Event-Driven Design
File system watchers detect new HLS segments, triggering AI processing and real-time WebSocket notifications.
Scalable Processing
Celery workers handle AI tasks with configurable queues for real-time vs batch processing modes.
Cloud-Native
Kubernetes manifests for local (KIND) and production (GKE) deployment with easy environment switching.