Home / All features / 3D player pose

● live in a club

3D player pose

Several court cameras see a player from different sides, and the court server builds their pose in three dimensions: arms, torso and legs in every rally. No marker suits or sensors — in normal play, for all four players at once.

3D view of court 4 with an orbiting camera: player poses and the ball computed from the court cameras (a training session)

What it is and why

One camera sees a player flat: you cannot tell whether the torso is turned, how much the knees bend, where the racket is relative to the body. When several calibrated cameras see the player from different sides, the pose can be rebuilt in 3D — like a biomechanics lab, but on an ordinary court during a game.

3D poses are the base of technique analysis: the player's stroke is compared with a reference movement, angles are measured and you see where the stroke breaks down. The video on this page is a real 3D scene of court 4: people's poses and the ball computed from the court cameras.

What it does

3D player pose

  • 3D body pose from several cameras, no sensors on the player
  • All four players at once
  • The basis for technique and movement analysis
  • Computed on the court server in the club
What the court needs

An AI Padel smart court: 4 overview cameras and a court server with a GPU.

How it works

How it works

  1. Cameras from different sides

    Overview cameras film the court from several angles at the same time.

  2. Joints in the frame

    A neural network finds the player's joints — shoulders, elbows, wrists, hips, knees, feet — in every camera frame.

  3. Building 3D

    Using the camera calibration the court server combines joints from several cameras into one pose in metres on court.

  4. All four players

    Poses are built for every match player, tied to time and rally.

  5. Computed after the match

    The heavy computation runs on the court server with a GPU — including at night when the court is free.

  6. Technique analysis

    Poses feed technique analysis: comparison with a reference, angles and tips for coach and player.

  7. 360° view

    On the terminal you can rotate the 3D court scene by touch and look at it from every side.

Ball in 3D: the trajectory from the court 4 ball cameras on a court diagram — three rallies, real time and ×0.5 slow motion

What you will see

What you will see

Player

In technique analysis — your pose and the reference movement over the video.

Coach

A student's movement in 3D: torso, arm and leg angles in the stroke.

Club

A 3D court scene with poses and the ball — for reviews and demos.

Limits, honestly

Limits, honestly

  • If a partner, the net or another player hides the player, the pose in those frames is rebuilt worse or not at all.
  • Several calibrated cameras and a court server with a GPU are required.
  • Poses are computed after the match, not during play.
  • This is a video estimate, not a lab marker-system measurement.

Who it helps

Who it helps

Players

Technique analysis from real play, not a drill on the spot.

Coaches

An objective picture of a student's movement: angles and stroke phases.

Clubs

Technology ordinary courts do not have — for the club's school and coaches.

API

API

The same data in your systems through the open API.

GET /v1/cv/match/{match_id} # match data from the cameras

API documentation →

Questions and answers

Questions and answers

Do players need marker suits?

No, normal sportswear.

Where is it computed?

On the court server in the club; video does not leave the club for this.

How many players at once?

All four match players.

When are poses ready?

After the match; the heavy computation may run at night when the court is free.

What does it give a player?

Technique and movement analysis from real play.

How many cameras are needed?

Several overview cameras from different sides, calibrated to the court.

Why does a pose sometimes jump?

When the player is hidden, fewer cameras see them and the estimate is worse.

Is the ball shown in 3D?

Yes, the 3D court scene also shows the ball path computed by the ball cameras.