jhwang7628 / monotrack

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.

Installation

1. Court Detection

See the court-detection folder for further instructions.

2. Pose Estimation

We use MMPose and MMDet for pose detection. See here for MMPose and here for MMDet.

3. Shuttle Tracking: Install our modified TrackNet

We use a modified TrackNet for shuttle tracking. See the modified-tracknet for further instructions.

4. Data Analysis / Match Statistics

Install our Python package.

cd python/ai-badminton/
pip install --user -e .

5. Prepare the dataset for hitnet

cd setup
python setup.py

6. Training HitNet

Run through the cells of modified-tracket/train-hitnet.ipynb.

7. Running our pipeline

See ai_badminton.pipeline_clean (i.e. pipeline_clean.py in python/ai-badminton/src/).