YUMI-TRACK

A web app that scores kyudo form from webcam video with pose estimation — the seed of Kyudo AI.

Period
2025-05 – 2025-07
Role
Developer and lead author
Category
Web
Status
Archived
Stack
Python, Streamlit, MediaPipe, NumPy

A web app I built in Python alone between May and July 2025. A PC and a webcam film the archer, and MediaPipe pose estimation scores four measures: equal distance and the three crosses (sanjū-jūmonji: shoulders, hips and feet). Once it was finished, members of the Yuge College kyudo club tried it.

I presented this work as “Proposal and Web Implementation of a Kyudo Form Evaluation Method Based on Real-Time Pose Estimation” at the 2025 Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers and received the Outstanding Presentation Award (publications).

A fixed PC and webcam are hard to bring into a dojo, and the app could not record hits. Rebuilding it as a smartphone app to solve this became Kyudo AI.

← All projects