This project started as an Urban Futurism competition idea. We wanted to build a smarter CCTV style system that can understand what is happening, but still respect privacy. I built the core vision pipeline using YOLO for real time object detection and OpenCV for tracking and region logic.
The fun part was privacy. Instead of dumping raw faces everywhere, we added privacy filters so the feed stays usable for safety, but sensitive parts are protected. We experimented with generative AI based masking and blur styles so it looked clean and did not break the scene context.
We ranked 7th overall. For me, this was a super cool intro to building end to end computer vision systems that actually run in real time, not just in a notebook.