Rock-Paper-Scissors Hand Gesture Recognition

Project Structure

my_rps_dataset/

Core dataset management and preprocessing tools:

• record_rps_video.py - Record webcam videos of hand gestures
• preprocess_and_export_landmarks.py - Extract MediaPipe hand landmarks from videos
• create_video_splits.py - Generate train/test/validation dataset splits
• features/ - Exported landmark coordinates in .npz format
• features_angles/ - Hand joint angle features derived from landmarks
• features_coords/ - Processed 3D coordinate features
• landmarks/ - Raw landmark data extracted from videos

main/experiments_dtw/

DTW-based classification experiments:

• dtw.ipynb - Dynamic Time Warping classification experiments
• check.ipynb - Data validation and analysis

alignment/

Video processing and data alignment tools:

• video_frames_to_img.py - Extract individual frames from videos
• data_check.ipynb - Validate landmark data at cropped video frames

Getting Started

1. Set up the environment:

   python -m venv rps_venv
   source rps_venv/bin/activate  # On Windows: rps_venv\Scripts\activate
   pip install -r my_rps_dataset/requirements.txt

2. Record gesture videos:

   cd my_rps_dataset
   python record_rps_video.py rock
   python record_rps_video.py paper
   python record_rps_video.py scissors

3. Extract landmarks and features:

   python preprocess_and_export_landmarks.py

4. Create dataset splits:

   python create_video_splits.py --data_dir features/ --out_file splits.json