Automated Hit-frame Detection for Badminton Match Analysis

This repo contains official implementation of Automated Hit-frame Detection for Badminton Match Analysis. The main contributions of our work are to:

• Utilize the interchange of shot angle to trim rally-wise frames from raw badminton video.
• Proposed a novel transformer that predicts shuttlecock direction sequences based on player keypoint sequence.
• Designed and developed the first-ever automated hit-frame detection tool to bridge the gap between raw badminton videos and analyzable data.

Paper: https://arxiv.org/abs/2307.16000

Checkpoints: https://drive.google.com/drive/folders/1v-uejba2ljNRUPaSAR-9u-a9GY1mD4pu?usp=sharing

Datasets: https://drive.google.com/drive/folders/1oZf7U5qx70YuNZX5Df6y1vJ_WwrMEprD?usp=sharing

Environment Requirements

opencv-python == 4.7.0.72
python == 3.8
scikit-learn == 1.0.2
torch == 2.0.1
torchvision == 0.15.2
yaml

YAML Parameters

- model:
  - sacnn_path: Path to SA-CNN's weight                               Default is './models/weights/sacnn.pt'.
  - court_kpRCNN_path: Path to Court Keypoint-RCNN's weight           Default is './models/weights/court_kpRCNN.pth'.
  - kpRCNN_path: Path to Keypoint-RCNN's weight                       Default is './models/weights/kpRCNN.pth'.
  - opt_path: Path to transformer model's weight                      Default is './models/weights/OPT_16_head_dp.pt'.
  - scaler_path: Path to data scaler's file                           Default is './models/weights/scaler.pickle'.
- sa_queue length: Length of the sa_queue                             Default is 5.
- video_directory: Directory with unresolved videos                   Default is '../videos'.
- video_save_path: Directory to store resolved videos                 Default is '../outputs/videos'.
- joint_save_path: Directory to store player joints and frame info    Default is '../outputs/joints'.
- rally_save_path: Directory to store rally-wise info                 Default is '../outputs/rallies'.

Run the Code

python main.py

Options

--yaml_path        STR      Path to AI Coach setting yaml.     Default is "../configs/ai_coach.yaml".