11 segments · 8 disciplines · measured overlay
Loading capture data
The stimulus labels (loaded drive, braking, peak impact, unloading) describe the mechanical event occurring at that instant, derived from the measured force vector. They are descriptions of loading, not claims of physiological outcome. Skeletal and muscular adaptation are longitudinal responses measured over weeks to months by clinical methods; nothing observable in a few seconds of video can evidence them, and no such claim is made here.
The Laso load figure is a direct calculation, not an estimate of benefit: for a belt of mass m carried at the centre of gravity, the additional force the athlete must produce is m × |a + g|, where a is the measured centre-of-mass acceleration. Accumulated mechanical work attributable to the belt is the time integral of that force against centre-of-mass speed; the kilocalorie figure converts that work at an assumed 25% muscular efficiency, a standard first-order assumption, and represents the belt's added contribution only — not total session expenditure. Belt mass is user-selectable in the toolbar and every figure updates with it.
Source clips were recorded on phones across different venues and lighting conditions. Each was measured for mean brightness, saturation and contrast, then graded toward a common target using a per-clip correction, a filmic tone curve, light sharpening and a soft vignette. This is a presentation treatment applied for legibility and consistency; it alters no measurement, because every value shown was computed from the original ungraded footage. Audio is removed.
The wireframe drawn on the footage is the measured skeleton itself, plotted in image coordinates at the position the pose estimator reported for each frame — it is not drawn by hand or fitted by eye. Bones are hidden when the estimator's confidence for a landmark falls below 0.3, which is why limbs disappear during occlusion rather than being guessed at. The chrome band at the hips marks the belt position derived from the two hip landmarks. The force arrow originates at that band and points along the estimated ground-force vector, scaled to the subject's size in frame.
Thirteen anatomical landmarks are extracted from each source clip at 30 frames per second using a convolutional pose-estimation model, producing three-dimensional joint coordinates in metres. Two segments were captured with a prior instrumented rig and converted to the same schema; their recorded flight time independently matched this pipeline's computed value to within 0.07 s.
Ground reaction force is derived from the second time-derivative of centre-of-mass position, expressed in body-weight units as 1 + a/g. The lateral component reported as cut force is the horizontal magnitude of that vector. Flight is detected as sustained free-fall acceleration combined with contact-point elevation, using ankles or knees as the contact reference; during flight, ground force is set to zero, since a body without ground contact produces none.
Force is a second time-derivative, so any clip recorded at a non-standard frame rate would yield invalid magnitudes. Each segment was therefore checked against physical plausibility. One segment — the receiver cut and catch — was identified as slow-motion source footage: its computed body speed of 2.9 mph is inconsistent with the movement shown. That segment is retimed for natural playback and its absolute force and speed values are withheld rather than published at an unverified scale; direction of loading remains valid. A second segment was independently validated instead: a recorded jump elevation of 31 inches implies a flight time of 0.801 s by projectile motion, and the measured flight time was 0.801 s.
Every movement shown is a real recorded performance. No motion is authored, retargeted, or idealised. Object positions — tennis ball, basketball — are colour-tracked from the source footage frame by frame, not animated.
IN MOTION