From raw footage to decision support.

A broadcast feed, the tactical camera behind the stand, or a phone on a tripod — the pipeline starts from ordinary video. No wearable sensors, no vests, no markers: players carry nothing, and nothing on the pitch changes.

Which footage each product works from — full match video for clubs, a single phone camera for individual athletes — is laid out on the solutions pages.

Detection locates every player in the frame; tracking holds each identity from one frame to the next — built to stay locked through occlusions, collisions, and the crowd of a corner kick. In the window above, every cycle re-runs detection on a fresh synthetic formation: the sweep passes, the boxes snap on, the confidence values count up and tick.

Each tracked player becomes an articulated skeleton: 21 joints, 25 times per second — 462 joints at once on a full pitch. Pose is estimated in the image and lifted from 2D to 3D, which is why a single ordinary camera is enough.

Above the skeletons sits a self-supervised foundation model of movement: it turns joint trajectories into a shared representation that can be compared, searched, and scored — the layer every capability on this page draws on.

• The 21-Joint Skeleton
• Foundation-Model Training
• Pose EstimationPrototype

The window above is lifted straight from the product: the movement-search console, one of the working windows in the full product suite in Preview.

Skeleton data compounds. Movement signatures give every player a baseline of how they sprint, cut, and land. Movement search finds every recurrence of a pattern across a season — queried by example, not by tag. Tactical fingerprinting and clustering read pressing patterns and team shape from the same joints. And when a player drifts from their own baseline — stride asymmetry creeping upward — that drift surfaces as an early injury-risk signal and, on the roadmap, feeds a squad readiness board.

What comes out is decision support in concrete form: recommendations, indicators, and flagged clips. They reach two product surfaces — the club console of Team Intelligence and the athlete’s phone in Personal Performance — and a second channel: AI agents that query the movement knowledgebase and answer with the evidence linked.

ResearchRoadmapForward-looking signals are labeled: injury-risk indicators are a research program; the readiness board is on the roadmap.