sagsan/golf-swing-analyzer
0
โณ Golf Swing Analyzer & Coaching Dashboard
An end-to-end Computer Vision and Machine Learning application that validates golf swing videos, pinpoints 8 key swing milestones using a Bidirectional LSTM, performs 2D biomechanical analysis, and provides synchronized pro comparison overlays and interactive kinematic charts.
๐ Key Features
- ๐ค Gatekeeper Validation: XGBoost binary classifier (99.0% accuracy) rejecting non-golf videos and long clips (>60s).
- ๐ง BiLSTM Milestone Locator: Predicts 8 chronological swing events (Address, Toe-Up, Mid-Backswing, Top, Mid-Downswing, Impact, Mid-Follow-Through, Finish) with an overall MAE of 2.66 frames.
- ๐ Biomechanical Scorecard: Measures Spine Tilt loss, Lead Arm Straightness, Hip Sway Ratio, and Foot Weight Transfer.
- ๐ฅ Side-by-Side Pro Comparison: Dynamic pro matchmaker against 16 pro reference models with synchronized speed control.
- ๐ Kinematic Charts: Interactive Plotly plots comparing user swing kinematics against matched pro trajectories.
๐ฆ Assets & Models
- ML Models: sagsan/golf-swing-analyzer-models
- Dataset & Presets: sagsan/golf-swing-analyzer-dataset
Built with OpenCV, MediaPipe, TensorFlow/Keras, XGBoost, and Streamlit.
