From Pixels to Points: Inferring Archery Scores with Computer Vision
summary
A working pipeline infers arrow tips and scores from photos with 3,500 annotated arrows and about 504 ends, using YOLO oriented detection plus a second pose model to resolve tip direction, then applies homography to face‑on coordinates to turn pixels into millimetres. Training runs in hours, detection hits ~31/32 arrows with rare false positives, and the system deploys to a Raspberry Pi for on‑range inference while failing most under harsh shadows and very tight clusters.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 4:01Limitations of scoring apps
- 7:32Training data for arrows
- 9:37Arrow tip detection
- 13:51Pixel to target mapping
- 16:40Full scoring pipeline
- 19:07Raspberry Pi deployment
- 21:12Future improvements
- 24:03Q&A
related talks
36m"Lights, camera, AI!" - Automated sports videography
40mThe eye of the Python, an eye tracking system. From zero to...
21mLeveraging Python for Real-Time Image processing
23mBirding with Python and Machine Learning
24mImage recognition and camera positioning with OpenCV. A tourist guide application.
31mYin Yang Ranch: Building a Distributed Computer Vision Pipeline using Python, O...
15mFOCAL Point: Utilizing Python, R, and Shiny to Capture, Process, and Visualize Motion
20mGuide to your own artificial intelligence application in 3 easy steps