High-load Real-time AI Image Processing System for Media Art
Technical Director
The problem I solved
To connect AI-processed data with Unity content in real time for a media artwork, I designed and built a distributed data-processing pipeline.
ENGINEER · STRUCTURE
Turning bottlenecks into a real-time processing system
I isolated bottlenecks in transmission, acquisition, and high-load processing, then connected Node.js, Python, and Unity into one real-time flow.
How I built the system
- Image data processed 60 times per second
- Node.js · Python · Unity · WebSocket
- Python multiprocessing reduced image-acquisition latency from 206ms to 13ms
- Multiple Raspberry Pi clusters distributed high-load processing
ARTIST · EXPERIENCE
Carrying processed data through to the artwork experience
As technical director, I worked with artists to build the production foundation that carried real-time data into Unity content.
How I connected it to the experience
- Participated in an artwork creation project with artists as technical director
- Integrated AI-processed data with Unity content in real time
- Key outcome
- I designed and built a real-time pipeline and Unity integration logic to process image data 60 times per second.
- My approach
- I treat performance as part of the full production flow, designing the path that carries data into Unity content in real time.
- Metric definition
- The 13ms figure is image-acquisition latency, not AI inference or end-to-end pipeline latency.