Identity from Movement

Scouts recognize a familiar player from the stands by stride and posture long before any jersey number is readable. Identity from movement is the research question behind that intuition: is movement style — gait, acceleration habits, coordination timing — distinctive enough to re-identify a player from skeleton data alone?

The starting evidence is real: at proof-of-concept scale, movement signatures separate players within a match using no appearance features at all — no jersey color, no face, no venue. Whether that separation holds across weeks, kit changes, cameras, and physical development is exactly what this research direction exists to measure.

The mechanism under investigation is signature matching. A tracked segment aggregates into a distribution in signature space; candidate identities are ranked by distance to known player profiles; a person confirms. Because signatures are invariant to camera and body size by training design, the comparison targets how a player moves rather than how they look. Nothing new is invented at the representation layer — this direction stands on movement signatures and asks one additional question of them.

Scouting in the wild is the motivating scenario: footage where identity metadata is missing or unreliable — youth tournaments filmed on a phone, unfamiliar leagues, crowded frames where numbers stay hidden. If re-identification survives evaluation, a scout could follow one player across such footage by movement alone, and archives without clean rosters would become searchable by player.