Mobile Jump Height Estimation

Problem

Accurate vertical-jump measurement has traditionally required force plates or motion-capture rigs — equipment far outside the reach of most athletes, coaches and physical therapists. Smartphone cameras, by contrast, are already in nearly everyone's pocket.

Why it matters

If a single phone camera can approximate lab-grade jump measurement, biomechanical feedback stops being something only well-funded programs can afford. That has direct value for coaching, injury-recovery tracking and everyday athletic training.

Constraints

• Single monocular camera — no depth sensor or multi-camera triangulation
• Variable frame rate and lighting conditions across consumer devices
• Pose-estimation noise is worst exactly at the frames that matter most: takeoff and landing
• Needs to work with little to no manual calibration to stay practical for real users

Approach

The system tracks hip and ankle keypoints across frames using on-device pose estimation, detects takeoff and landing by identifying sign changes in vertical velocity, and converts the resulting flight time into an estimated jump height using standard projectile kinematics.

Experiments

• Comparing flight-time-based height estimates against a reference measurement method
• Testing sensitivity to frame rate to identify a minimum viable capture rate
• Evaluating robustness to camera angle and distance from the subject

Results

• Flight-time estimation is workable but highly sensitive to frame rate near the takeoff/landing boundary
• Camera angle tolerance is narrower than initially expected before error grows meaningfully

Lessons

• Keypoint jitter matters far more at takeoff and landing than mid-flight — that's where error budgets should be spent
• A biomechanically-motivated model (flight-time kinematics) generalizes better across subjects than a purely learned regression would with this little data