Sports Vision · computer vision for the game

AI that watches the game and calls every shot.

ataraxy builds real-time computer vision for sports video. Point a camera at the court and every shot is detected, tracked and called as it happens: makes, misses, trajectories, scores. Live from a camera or from uploaded footage. Built on your infrastructure, owned by you.

01
The gap

Game film doesn't watch itself.

Every week, coaches and analysts burn hours tagging footage by hand. The stats arrive late, incomplete, or not at all.

The gap

Video is the richest data a sports organization owns, and most of it goes unwatched. Manual tagging is slow and expensive, so it only happens for the games that matter most. The big sports-analytics vendors will do it for you, but you pay per game, per season, forever, and the models, the data and the pipeline all stay theirs.

The winning move is a vision system of your own: cameras you already have, models tuned to your sport, an API your product and your staff can call. Every game watched, every shot counted, and the whole engine belongs to you.

02
What it does

Eyes on the ball, every frame.

One vision engine that detects the ball and the rim, follows the trajectory, and calls the outcome the moment it happens.

Shot detection

Every make and every miss, called live

Detection models trained for the court find the ball and the rim in every frame, and a shot classifier calls the outcome. Makes, misses and scores land in your feed seconds after they happen on the floor.

Trajectory tracking

It follows the ball, not just the frame

The engine tracks the full arc of every shot across frames, through hands, bodies and bad angles. That trajectory is what separates a real attempt from noise, and it is how the classifier knows what actually happened.

Live streaming

From camera to call in under a second

A streaming pipeline takes compressed video straight from the camera over WebSocket and processes it frame by frame as it arrives. No upload-and-wait: the game and its stats run on the same clock.

03
How it works

How the engine sees.

Stream, track, call. Live camera or uploaded footage, same brain.

01

Stream

Video comes in live from a courtside camera over WebSocket, or as full-game files through the batch queue. Same pipeline either way, so accuracy in the lab is accuracy on game night.

02

Track

Frame by frame, detection models find the ball and the rim while the tracker stitches detections into trajectories. Every arc toward the basket becomes a candidate shot with a full flight path.

03

Call

The classifier rules on every attempt: make or miss, plus running score. Results stream out as structured events your app consumes over the API, with annotated video when you want the receipts.

One pipeline. Live camera or uploaded footage, same brain.
04
Ownership

You own the engine.

Not a per-game subscription to somebody else's black box. We build it, you keep it.

Your infrastructure

Runs where you run

The engine deploys on your own cloud with GPU when you need throughput. Your footage is processed on infrastructure you control, and cleanup policies mean video never lingers longer than it should.

API-first

Plugs into your product

Everything the engine sees is available as clean, structured events over an authenticated API and WebSocket. Feed a live scoreboard, a coaching dashboard, a fan app, or all three at once.

No lock-in

No per-game meter running

A fixed build, then it is yours. No per-game fees, no per-season renewals, no vendor holding your own footage hostage. It keeps running whether or not we are in the room.

Your data

Your footage, your models, your stats

The models are tuned on your context and handed over documented, with an accuracy review workflow so your team can keep them sharp as conditions change. Nothing about how it works is a secret to you.

We stay available. We don't stay necessary.
05
Beyond basketball

Built for the court. Ready for any camera.

Basketball is where we proved it. The same engine learns whatever your cameras watch.

If a camera can see it, the engine can call it.

Shots at a rim are just events in a video stream: an object, a trajectory, an outcome. The same pipeline that calls makes and misses can call goals and saves, serves and faults, reps in a gym, vehicles in a lot, or safety events on a factory floor.

If you have cameras and questions they should be answering, that is exactly the kind of system we build: detection tuned to your world, real-time when it matters, delivered on your infrastructure and owned by you.

Put eyes on your game.

A 30-minute call. Bring the footage or the cameras you already have; we'll map what a vision engine can see for you, and scope a fixed price before you commit to anything.