01 / THE PROBLEM
The problem
Empty spaces should not be hard to find.
02 / WHAT I BUILT
What I built
Live vehicle detection turns a camera feed into parking availability.
03 / THE HARD PART
The hard part
Vehicle detection alone was not enough. Detections had to map reliably to individual parking regions while the video loop kept up.
04 / ARCHITECTURE
Architecture
YOLOv5 and PyTorch detect vehicles. OpenCV handles the video pipeline, while polygon-based regions represent individual parking spaces.
05 / RESULT
Result
110+ spaces monitored. Smart Campus winner.
06 / WHAT I’D DO DIFFERENTLY
What I’d do differently
The most useful next step would be testing across changes in lighting, camera position, and occlusion, with clear measures for slot-level reliability.