CASE STUDY · REAL-TIME IMAGE PROCESSING
I built a real-time image-processing system and cut image-acquisition latency from 206ms to 13ms.
As technical director, I worked with the artists and owned the technical scope from planning through delivery. I built a pipeline for 60 image-data cycles per second and used Python multiprocessing to reduce image-acquisition latency from 206ms to 13ms.
Problem-solving steps
These steps are an explanatory sequence, not a claim about the chronological order of the original decisions.
01 · CONTEXT
I was responsible for the full technical scope of the media-art production.
As technical director, I worked with artists on the media-art production and was responsible for the full technical scope from planning through delivery.
Key detail · Role
Technical director · full technical scope from planning through delivery
02 · REQUIREMENT
Image data had to be processed 60 times per second.
To connect AI-processed data with Unity content in real time, I designed for a requirement to process image data 60 times per second.
Key detail · Recorded condition
Process image data 60 times per second
Derived reference · Derived value
If 60 cycles per second are assumed to be evenly spaced, each cycle is about 16.7ms. This is not evidence of actual input cadence, a processing budget, or end-to-end latency.
03 · IMPLEMENTATION
I built real-time image transmission and processing and integrated it with Unity.
I used Node.js, Python, Unity, and WebSocket to implement real-time image processing and Unity content integration.
Key detail · Technologies
Node.js · Python · Unity · WebSocket
Interpretation note · Boundary
The actual network topology and AI model are not recorded
04 · INTERVENTION
I applied Python multiprocessing to image acquisition.
I applied Python multiprocessing to reduce image-acquisition latency.
Key detail · Applied technique
Python multiprocessing
Interpretation note · Boundary
Alternatives, tradeoffs, and the original selection process are not in the current record
05 · DISTRIBUTION
I distributed high-load processing across multiple Raspberry Pi clusters.
I distributed high-load processing across multiple Raspberry Pi clusters.
Key detail · Implementation
High-load distributed processing with multiple Raspberry Pi clusters
Interpretation note · Boundary
Device count, topology, and direct causality for latency improvement are unverified
06 · OUTCOME
I reduced image-acquisition latency from 206ms to 13ms.
The two values are image-acquisition latency before and after Python multiprocessing.
Key detail · Measurement
Image-acquisition latency 206ms → 13ms
Derived reference · Derived value
The two recorded values correspond to an approximately 93.7% reduction.
Interpretation note · Boundary
Measurement setup, sample count, and system-wide end-to-end latency are not in the record
What I built and achieved
I addressed the performance problem with a real-time pipeline, multiprocessing, and distributed processing. Where the measurement scope is unavailable, I do not extend the claim beyond the recorded result.
What I implemented
- ✓Designed and built a real-time pipeline to process image data 60 times per second.
- ✓Reduced image-acquisition latency from 206ms to 13ms with Python multiprocessing.
- ✓Implemented high-load distributed processing with multiple Raspberry Pi clusters.
Limits of the figures and interpretation
- —The exact cause of the 206ms bottleneck and alternatives considered at the time
- —Any interpretation of 13ms as end-to-end pipeline or AI-inference latency
- —The exact Raspberry Pi count, topology, or direct causal link to the latency change
My role and project outcomes
As technical director, I owned the work from planning through delivery and led the real-time pipeline implementation and performance improvement.
High-load Real-time AI Image Processing System for Media Art
Technical Director
As technical director, I worked with the artists from planning through delivery. I designed and built a real-time pipeline to process image data 60 times per second and implemented the core logic connecting AI-processed data with Unity content in real time.
Roles and responsibilities
- Technical Director: Handled the entire technical part from planning to delivery.
- Backend/AI/3D Graphics: Designed and implemented a real-time image transmission and processing pipeline utilizing Node.js, Python, and Unity.
Key outcomes
- Secured Real-time Performance: Shortened image acquisition latency from 206ms to 13ms utilizing Python multiprocessing.
- Architecture Optimization: Implemented high-load distributed processing by building multiple Raspberry Pi clusters.
Technical environment
OS: Windows, Ubuntu / Communication: WebSocket / Languages: Node.js, Python / Engine: Unity / Tools: VSCode, Git