Research
Automated Race Timing System
A computer-vision race-timing system that detects athletes, reads their race bibs and tracks them across frames to time events automatically.
Result
Over 99% bib-reading accuracy in validation, at 60+ FPS on a consumer GPU.
01
My part
This was a university group project. I designed and built the computer-vision pipeline end to end.
02
How it works
- YOLOv11 detects athletes and bibs in each frame.
- ByteTrack follows each detection through the scene over time.
- Bib crops go through OCR (EasyOCR / PaddleOCR) to read the number.
- Readings are resolved into athlete identities and timing results.
- Results go through a REST API into a Django app, where organisers can check any uncertain reads by hand.
03
Performance
More than 99% bib-reading accuracy on the validation set, running at over 60 frames per second on a suitable consumer GPU.