Machine learning · Computer vision · Mobile app
BIRDIE: Real-Time Badminton Form Evaluation
An in-progress mobile coaching tool!
BIRDIE helps players review their stance when practicing their badminton form.
The project
Project focus
I am building BIRDIE with a badminton teammate to make form analysis more accessible to newer players. Instead of requiring a specialized camera installed at a particular court, the project explores how a player’s own camera can support portable feedback and practice.
Data and modeling
We created and labeled a custom dataset of badminton form examples. The form-detection work combines MediaPipe pose landmarks, OpenCV video processing, calculated joint angles, logistic regression, and convolutional neural networks. These experiments helped us translate movement into features that a model can compare and classify.
Current direction
The project is still in development. Current work focuses on strengthening the dataset, improving model behavior across different camera angles, connecting the analysis pipeline to the mobile interface, and turning technical classifications into feedback that feels useful on the court.
Form-detection Loop
-
01
Capture
Record a player’s movement through the mobile camera.
-
02
Map
Use pose landmarks and joint angles to describe badminton form.
-
03
Classify
Evaluate form with computer-vision and machine-learning models.
-
04
Guide
Turn model output into approachable feedback for newer players.
Development archive
Project Notes
Pose tracking, angle calculations, building the dataset, CNN training, app integration, and crashouts :p
Prototype demo