Simulation
Weightlifting Force Estimation with Motion Magnification
A motion magnification technique for objects undergoing large rigid-body motion amplifies sub-pixel elastic barbell deformations, extending the video-based force estimation system to lower training weights where barbell flex is invisible to the naked eye.
Links & Resources
Background
The weightlifting force estimation system fuses video-based pose tracking with a flexible multibody model of the barbell to infer ground reaction forces and barbell forces from standard footage. The core mechanism relies on matching the elastic deformations of the digital and physical barbell twins: when the bar bends under load, that deformation encodes information about the forces applied. This makes accuracy weight-dependent — at lower training weights, where barbell flex is barely perceptible, the estimator has too little deformation signal to work reliably.
Motion Magnification Under Large Rigid-Body Motion
To extend the system's working range, a variant of phase-based motion magnification was integrated — one specifically adapted for objects undergoing simultaneous large rigid-body motion. Classical phase-based motion magnification is designed for roughly stationary scenes; applying it directly to a barbell moving through space requires separating rigid-body translation from elastic deformation at the sub-pixel level. The adapted algorithm does this separation, revealing microscopic flexing that the unaided eye cannot detect, and providing the fusion estimator with enough resolution to synchronise the digital and physical twins even at low loads.
Demonstration
The approach is demonstrated on a one-arm power clean with a total barbell mass of 60 kg. At this weight, elastic deformation is invisible without magnification. With a magnification factor of 100, the algorithm clearly reveals the microscopic bending to which the digital twin is matched, and the estimator produces stable force traces throughout the movement.
The body-parts segmenter occasionally confuses an arm for a leg during the lift — a known limitation: it was trained on clean-and-jerk and snatch footage rather than single-arm movements.
Note on AI's Role
AI plays a narrow but essential part in this pipeline: its only function is to generate masks for the barbell and weight plates, isolating them from the rest of the frame. The force estimation itself is grounded in image processing mathematics, vibration theory, and the physics of elastic bodies moving through space.