Ball-tracking produces a projected path for a delivery that was stopped by a pad. The projection is a physical model built from measurements, and understanding it explains where its uncertainty lies.
Several cameras see the same ball
High-speed cameras are placed at surveyed positions around the ground, each recording many frames per second. Software identifies the ball in each frame by its shape, colour and movement against the background.
Because the camera positions are known precisely, a ball seen simultaneously by two or more of them can be located in three dimensions. The process repeats for every frame of the delivery.
The result is a dense series of points describing where the ball actually was, from release until it struck the batter. Nothing in this stage is predicted; it is all observed.
A curve is fitted through the points
The observed points are noisy, so the system fits a smooth trajectory through them rather than joining them directly. The curve is constrained by the physics of a spinning ball moving through air.
Fitting also allows the system to infer speed, swing and the deviation caused by seam or spin. Those quantities fall out of the shape of the curve rather than being measured separately.
Confidence in the fit depends on how many frames captured the ball cleanly. A delivery obscured by the umpire or the bowler for part of its flight yields fewer usable points.
The bounce is the difficult moment
When the ball pitches, it loses speed and changes direction in ways that depend on the surface, the seam angle and the amount of spin. This is where prediction becomes genuinely hard.
The system measures the trajectory after the bounce as well as before it, so the bounce is observed rather than assumed. That post-bounce segment anchors the projection that follows.
The shorter the gap between pitching and impact, the less post-bounce flight there is to observe. Deliveries that strike the pad very soon after bouncing carry the widest uncertainty.
Projection extends the observed path forward
From the point of impact the software continues the fitted curve to the stumps, applying the deceleration and deviation already measured in the visible portion of the flight.
This is why the projected segment is drawn differently from the observed one in graphics. The two parts have different status: one is a record, the other a calculated continuation.
Competitions publish a tolerance representing the system's accepted error, and decision protocols use it. A projection clipping the stumps marginally is treated as inconclusive rather than as a wicket.
Calibration is checked before play
Cameras are aligned against surveyed reference points on the ground, and the stumps and creases are measured into the model. Any camera knocked during the day invalidates that alignment.
Operators run verification deliveries and compare the system's output against known positions. A failed check means the technology is withdrawn for that match rather than used with a caveat.
The protocol matters because the output looks authoritative. A clean graphic conveys certainty the underlying measurement may not have, which is why the calibration record sits behind every review.
