How does ball tracking work in cricket?

What is ball tracking in cricket?

Ball tracking turns video of a delivery into the ball's path through the air. A computer finds the ball in each frame, works out where it was on the pitch and how high, and joins those points into one smooth flight. Everything on a tracking graphic comes from that flight.

Once you have the flight, the numbers follow. Release speed is how fast the ball is moving just after it leaves the hand. Length is where it hits the pitch, and line is where it's heading relative to the stumps. Swing is the curve in the air, spin shows as the change of direction off the pitch, and the same path carried forward is what a review uses to say whether the ball would have hit the stumps.

How does ball tracking work with several cameras?

Put several cameras around the pitch, each looking from a different direction, and each one tells you which direction the ball is in. Where those directions cross is where the ball is, in three dimensions.

It's the same idea as fixing a point on a map from two places. It only works if the system knows exactly where each camera is, and that much kit is a lot to set up at a club net.

How can one phone track a ball in 3D?

One camera sees the ball as a dot on a flat picture. It knows which direction the ball is in, but not how far away. The phone fills that gap with two things it does know: the size of the pitch and stumps, and the way a ball has to travel through the air.

First, the pitch. The stumps stand 22 yards apart. Once the phone has found both sets in the picture, it knows where it is standing and which way it is pointing.

Then, the flight. A ball near the phone and a ball further away can sit on the same spot on the screen, but a real ball can't jump between them from one frame to the next. It leaves the hand, drops under gravity, bounces once and carries on. Only one path like that matches where the ball appeared in every frame, and that path is the ball's flight in 3D.

How does a cricket ball tracking app work, step by step?

In four steps. A neural network trained on cricket footage finds the ball in every frame. The app recognises the pitch and stumps and builds a 3D scene of the net. A physics simulation recreates the ball's 3D trajectory. Bowling speed, line, length, bounce, swing and spin are read from that trajectory.

Finding the ball

A neural network is a program that learns to recognise things from examples. Shown enough cricket footage with the ball marked, it learns what a ball looks like in flight: small, fast, often blurred, against nets, grass, kit and sky. In each frame it reports where on the screen it thinks the ball is.

Building the 3D scene

The stumps stand 22 yards apart, so once the app has found both sets it can work out where the phone is, which way it points and how many metres each part of the picture covers. That turns a flat video into a 3D scene of your net.

Recreating the trajectory

A physics simulation knows how a ball moves: gravity pulls it down, the air slows it, and it bounces once off the pitch. It searches for the one path from release to the batter that lands on every spot where the ball was seen.

Reading the numbers

Every number comes from that one trajectory. Release speed is distance along the path divided by the time it took, read as the ball leaves the hand. Length is where it hit the pitch, line is where it was heading against the stumps, swing is how far it moved sideways in the air and spin is read from how far it changed direction off the pitch.

How does ball tracking predict where the ball would have gone?

It carries the 3D trajectory on past the point where the ball was stopped. For an LBW, the physics simulation keeps the ball moving from where it hit the pad on to the stumps. The further the path has to be carried, the less certain the prediction is.

A path carried half a metre is far more certain than one carried three metres. That's why a review needs a margin: when a check is too close to call, the on-field decision should stand.

YORKER AI uses the same idea for DRS on your phone. You make the call first, then the app checks pitching, impact and wicket from the tracked path, and replays the review in 3D.

What can trip up single-camera tracking?

Mostly the setup. A pitch shorter than 22 yards, a phone that moves, poor light, or a ball hidden at release or lost at the far end all leave the tracking less to go on. A good system notices when it hasn't seen enough and tells you, instead of guessing.

• A short pitch. The phone assumes the stumps are 22 yards apart, so on a shorter pitch every distance, and every bowling speed, comes out wrong.
• A phone that moves. If it moves after you've lined up the stumps, it no longer knows where the pitch is.
• Not seeing enough of the flight. The more of the flight the camera sees, the clearer the answer.
• A small, fast ball. At the far end a cricket ball is only a few pixels across, and in poor light it blurs.

Our guide to phone tripod setup in the nets covers how to avoid each one.