Hockey Goal Detector
In progressA rink-side computer vision project that uses a YOLO model to detect goals and trigger an old Bud Light goal light through a small hardware bridge.
- Role
- Builder
- Timeline
- 2026 – present
- Status
- In progress
- Stack
- YOLOPythonOpenCVDockerGPIO
Goal
The project is simple in the best way: point a camera at a hockey net, detect when the puck crosses the line, and fire an old Bud Light goal light like it belongs in a basement rink broadcast booth.
The interesting part is making the loop dependable. A YOLO model can find candidate puck and goal events, but the system still needs confidence thresholds, frame-to-frame smoothing, false-positive handling, and a hardware trigger that fails calmly instead of strobing every time someone skates through the crease.
Architecture
[ Camera ] -> [ YOLO detector ] -> [ event filter ] -> [ relay / controller ] -> [ Bud Light goal light ]
The detector watches for goal-line events, the filter turns noisy frame predictions into a single clean event, and the hardware bridge handles the old goal light without coupling the model loop directly to the relay.
Current Focus
- Collecting representative clips from different angles and lighting conditions.
- Tuning the YOLO confidence threshold so it catches real goals without firing on near-misses.
- Building a small trigger service that can run locally and keep the hardware side isolated.
What Comes Next
The next milestone is a repeatable test set: clips with real goals, saves, traffic in front of the net, and no-puck motion. Once that is stable, the goal light becomes the fun part instead of the debugging surface.