This codebase contains the system described in MonoTrack: Shuttle trajectory reconstruction from monocular badminton video. MonoTrack is an end-to-end system for reconstructing 3D and 2D trajectories from broadcast-style badminton videos.
See the court-detection folder for further instructions.
We use MMPose and MMDet for pose detection. See here for MMPose and here for MMDet.
We use a modified TrackNet for shuttle tracking. See the modified-tracknet for further instructions.
Install our Python package.
cd python/ai-badminton/
pip install --user -e .
cd setup
python setup.py
Run through the cells of modified-tracket/train-hitnet.ipynb.
See ai_badminton.pipeline_clean (i.e. pipeline_clean.py in python/ai-badminton/src/).