01
Motion extraction
The backend turns ordinary phone video into structured movement data quickly enough to support real feedback.
- MediaPipe Holistic extracts body and hand landmarks frame by frame
- Supports front, side, and diagonal camera angles
- OpenCV overlay output makes the tracking inspectable instead of opaque
02
Phase segmentation
The system does not analyze the jumper as one blur. It breaks the motion into the moments coaches actually care about.
- Load, set, rise, release, and follow-through are detected automatically
- Knee angle, wrist height, and wrist velocity drive the segmentation cues
- Wrist flexion angle and speed become explicit flick signals
03
Feedback surface
The output is designed for correction, not novelty. The project translates raw motion into something a player or coach can act on after one upload.
- Flags issues like shallow knee bend, weak wrist snap, and short follow-through
- Returns keyframes, issues, confidence notes, and optional annotated video
- Grounds thresholds in biomechanics references instead of arbitrary heuristics