Audience Analytics with Computer Vision
A Python/OpenCV system measuring how people interact with advertising screens, with an associated IoT web application — AI work that predates the LLM wave.
Generative AI is one more capability — not a replacement for the fundamentals.
01Context
Digital advertising screens rarely know whether anyone is looking at them.
02Problem
Measure real audience interaction from a camera feed, and expose it through a web interface.
03Constraints
- Edge hardware
- Real-world lighting and crowds
- Privacy-conscious measurement
04My role
What I personally built
- Computer-vision interaction analysis in Python/OpenCV
- The associated IoT web application
What others owned
- Hardware, screens and commercial deployment were handled by Utrender
05Architecture
Architecture explorer
The business flow — what happens, in plain words.
01 / 03
Camera — Watches the space in front of the screen.
06System
Camera frames processed with OpenCV to detect and aggregate interaction events, sent to an IoT web application for monitoring.
07Challenges
- Robustness outside lab conditions
08Outcome
Delivered system and web application for Utrender.
09What I learned
Real-world data is the hardest part of any ML system — a lesson that carried straight into data engineering and LLM work.
10Stack
- Python
- OpenCV
- Computer vision
- IoT
- Web application