Computer Vision in Sport: How Hawk-Eye, Semi-Automated Offside and Tracking Data Work

Centre Court, 2007. Roger Federer and Rafael Nadal are deep into the fourth set of the Wimbledon final when a ball drops near the line. Hawk-Eye calls it in, by a single millimetre. Freeze the TV pictures, though, and the ball looks out.

Paul Hawkins, who invented the system, had an answer. The ball clipped the line by 1mm and then skidded on the grass, so the TV frame caught it after it had already travelled past the paint. A tennis ball can skid up to 10 centimetres when it lands. One broadcast frame can tell a very convincing lie.

That single point sums up computer vision in sport. The graphic looks like a photograph. It isn't one. And the gap between what the machine knows and what it shows you is where every argument starts.

Seven weeks into the 2026/27 Premier League season, these systems run through almost every match you bet on: goal-line technology, semi-automated offside, and the optical tracking that feeds live odds and xG models. Knowing how they work, and where they stop working, makes you a sharper punter.

Take away the branding and the principle is the same one your eyes use to judge distance. Point several synchronised high-speed cameras at the same patch of court or pitch from different angles. In every frame, software picks out the cluster of pixels that is the ball. Because at least two cameras see it from separate positions at the same instant, the computer can triangulate where it sits in three dimensions.

Repeat that frame after frame and you've got a trajectory. At Wimbledon in 2008 the system used ten cameras to photograph the ball in flight. "It then joins the dots," as Hawkins put it. Once the path is known, the software projects where the ball would have gone next and where it meets a pre-built model of the playing area: the lines, the stumps, the goal line.

That's why Hawk-Eye's output is described as the ball's statistically most likely path. What you see on screen is a model built from measurements, not a camera shot.

Frame rate is the unsung hero here. Take an illustrative ball travelling at 30 metres per second:

distance between frames = ball speed / frames per second

at  25 fps:  30 / 25  = 1.20 m between snapshots
at 200 fps:  30 / 200 = 0.15 m between snapshots

At broadcast-style frame rates the ball jumps more than a metre between pictures, so the exact moment it kissed a line or left a boot never gets captured. Faster cameras close that gap, and so do sensors inside the ball itself.

Hawk-Eye is a British invention. Engineers at Roke Manor Research in Romsey, then a Siemens subsidiary, developed it in 2000, with Paul Hawkins credited as the inventor. It was spun out as Hawk-Eye Innovations, based in Winchester, and made its broadcast debut on Channel 4 during England's Test against Pakistan at Lord's on 21 May 2001, analysing lbw decisions.

It spread quickly after that. Tennis tested it, cricket built a review system around it, and football eventually trusted it with goals. It's now used in more than 20 major sports, including snooker, badminton, rugby union, baseball and volleyball.

The money side is just as eye-catching. A business bought for £4.4m in 2006 went to Sony less than five years later, and the Japanese giant still owns it.

"You couldn't have a trophy that said, 'We are 99.76 percent certain that Roger Federer won.'"
— Paul Hawkins, inventor of Hawk-Eye

That was Hawkins answering a fair challenge. In 2008 the makers claimed an average accuracy of 3.6mm, roughly the width of the fuzz on a tennis ball. The ITF tested it against a 2,000-frames-per-second camera and found it comfortably inside the 5mm limit it sets for such systems. The advertised figure today is tighter again, at 2.6mm.

Notice the word average. Hawkins tuned the system to be sharpest where it counts, at the back edge of the line, with the error growing the further from it the ball lands. Even so, a 1mm "in" call sits inside a 3.6mm average error. The honest version of that 2007 decision is a probability, not a certainty, and that's exactly what Cardiff University's Harry Collins meant when he argued Hawk-Eye should show error bars. "There will always be errors," he said.

Hawkins' reply was that sport needs a verdict, and he's got a point. Tennis went for a clean answer. Cricket found a smarter one.

Cricket builds the doubt into the rules

In the Decision Review System, ball tracking plots where a delivery pitched, where it struck the batter and its projected path on to the stumps. When the projection is marginal, the on-field decision stands: umpire's call. Since the ICC's November 2017 rule change, a side keeps its review when an lbw appeal comes back as umpire's call.

That's error bars in all but name. The machine wins when it's clearly right, and the human decision survives when the data can't split in from out. We reckon it's the most intellectually honest use of tracking technology in any sport.

Wimbledon went fully electronic in 2025, dropping line judges after 147 years and handing every call to electronic line calling. The ATP adopted it across all its events that season, and the WTA at many of theirs.

The machine is only as good as the people running it, mind. In the fourth round on Centre Court, Sonay Kartal against Anastasia Pavlyuchenkova, an operator error left ball tracking switched off for a game. A point had to be replayed, and the All England Club took away the Hawk-Eye operators' ability to switch tracking off manually.

There were no player challenges and no video replays under the new set-up that fortnight. The system's word was final. That's a long way from the rows of 2007.

Football held out for years, then one moment at the 2010 World Cup forced its hand. In England's 4-1 defeat by Germany, Frank Lampard's shot hit the crossbar and bounced down over the line. No goal given. Sepp Blatter said FIFA would reopen the debate, and in July 2012 the IFAB approved goal-line technology.

The Laws permit it rather than require it, and the job is deliberately narrow. The rule itself hasn't changed: a goal counts only when the whole of the ball crosses the line between the posts and under the bar. FIFA's quality standards require a system to:

• Decide one thing only, whether a goal has been scored.
• Confirm within one second, via a vibration and visual alert on the referee's watch.
• Talk only to match officials, with nothing sent to anyone else.
• Pass an installation test, then a check by the referee before every match.

Hawk-Eye's version arrived in the Premier League in 2013/14, and every World Cup since 2014 has used goal-line technology. By September 2024 FIFA listed 144 stadiums with licensed systems, 135 running Hawk-Eye and nine using Vieww. Lampard-style ghost goals belong to another era.

Offside was always the tougher nut. It isn't one ball against a fixed line; it's 22 moving bodies, a moving ball and the precise instant of a pass. Semi-automated offside technology (SAOT) combines tracking cameras, computer-generated offside lines and, in some versions, a sensor in the ball. The "semi" is there because the VAR still reviews and confirms every decision.

It made its debut at the 2021 FIFA Arab Cup before the big stage in Qatar.

The sums show why no human could do this unaided:

29 body points x 50 readings per second = 1,450 readings per player per second
1,450 x 22 players                      = 31,900 readings per second (before the ball)

UEFA brought it in for the 2022 Super Cup and the 2022/23 Champions League group stage, and the 2023 Asian Cup in Qatar used it in all 51 matches. The Premier League switched it on on Saturday 12 April 2025, Match Round 32 of 2024/25, after non-live testing in the league and live use in that season's FA Cup. The league says it automates key elements of the decision to support the VAR, placing the virtual line more efficiently and generating the graphics fans see.

What it doesn't touch is Law 11. A toe offside is still offside. SAOT makes the line quicker and more consistent, but those tight calls aren't going anywhere. If you want the numbers on how video review shapes results, our guide to VAR's impact on football has them.

The offside cameras are just the visible tip of something far bigger. Genius Sports, which owns Second Spectrum, has been the Premier League's official tracking data provider since a partnership that began in 2019. Every Premier League and Championship ground carries the kit: around 28 to 30 cameras per stadium, mounted high up. According to Genius Sports' Michael D'Auria, they're actually smartphones, capturing up to 200 frames per second.

The goal is a digital twin of the match, a live 3D model of every player and the ball. Genius Sports says it now captures thousands of points on the surface of every player and the ball, hundreds of times a second. In the early days Second Spectrum tracked a single centre-of-mass point per player, in 2D, 25 times a second, and delivered it the next morning. Now the data lands in under a second. The family tree, D'Auria notes, runs back to missile defence and across to self-driving cars, and the offside system runs on exactly the same data.

That feed goes everywhere. Coaching and recruitment departments use it. So do broadcast graphics on Sky Sports and the Premier League's own Datazone, pressing and "between the lines" metrics, and the officials' SAOT.

Why tracking data makes xG sharper

Expected goals is a model's estimate of how many goals would be scored, on average, from a given situation, and StatsBomb rates it a better predictor of future performance than actual goals. Not all xG is equal, though. Since 2018 StatsBomb has recorded "freeze frames" of the goalkeeper and defenders at the moment of every shot, measuring how many bodies sit in the triangle between shooter and posts and how much of the goal is covered.

Its neat fix was mathematical. Old models counted blockers as whole numbers, so xG lurched when a defender shuffled a few centimetres across the edge of the triangle. Swapping hard edges for bell-shaped curves smooths that out:

old:  blocker = 1 if inside the shooting triangle, else 0      # cliff-edge jumps
new:  blocker = bell curve value between 1 and 0               # fades smoothly with position

Full optical tracking goes further still, giving models every player's position continuously rather than a single snapshot at the moment of the shot. If this side of the game is your thing, our expected threat (xT) guide shows how possession value gets built from the same raw material.

This is where the technology reaches your account. Most bookmakers don't build live prices from scratch. They buy official data feeds and odds from a small number of suppliers. Genius Sports sells low-latency data and odds APIs including official Premier League content, says it works with 400-plus leagues and 300-plus sportsbooks, and prices more than 600,000 events through its models and traders. Sportradar, official betting partner of UEFA, the Bundesliga, Wimbledon, the US Open and Roland Garros, sells player and micro-markets on the same principle.

The chain goes like this. Cameras and event feeds tell the model the state of the match: who has the ball, where, and under what pressure. The model turns that into probabilities, traders keep an eye on it, and the price updates. The speed and accuracy of that feed decide how quickly a market settles and how long it stays suspended.

Three practical consequences

You won't out-react the feed. Your TV picture or stream runs behind what's happening in the stadium. By the time you see a shot cannon off the post, the price has moved or the market has shut. The old courtsiding edge, punters in the stands betting quicker than the bookies' data, has been engineered out of existence.

Micro-markets live and die on data speed. Next corner, next goal kick, next throw-in: these markets only exist because event data now arrives quickly enough to price and settle them. They're fast and fun, and the margin is set by a model that sees more than you do. Our micro-betting guide explains how they're priced.

Decision tech shapes suspensions. Goal-line technology confirms within a second, so goal markets settle cleanly. Offside checks still wait for the VAR to confirm the line, and markets sit suspended until they do. Our guide to how VAR affects betting markets spells out what that means for in-play punters.

• Bet before kick-off, not against the feed. Pre-match is where your homework can beat a price. In-play, the supplier's data is quicker than your eyes every single time.
• Treat micro-markets as entertainment. Next-corner and next-throw markets are priced by machines watching thousands of data points a second, so keep the stakes small.
• Respect the line, whatever it says. SAOT draws offside faster and more consistently, but tight calls are still part of the game, so never count a goal as banked until the check is done.
• Use quality xG, not any xG. Models built on freeze frames and tracking data see blockers and space; cruder shot-only numbers can flatter a side or judge it too harshly.
• Stop pricing in ghost goals. Goal-line technology settles the over-the-line question inside a second, so that old source of chaos has gone from goals markets.
• In cricket, know the umpire's call. Marginal lbw shouts follow the on-field decision, so a not-out call carries real weight in tight moments.
• Love the tech, mind the operator. Wimbledon 2025 proved the person running the system is still part of the system.

Whatever the technology does, your discipline stays the same: fixed stakes, a set budget and no chasing in-play losses.

Related guides

• VAR in football: impact, controversy and statistics
• How VAR affects football betting markets
• Micro-betting explained
• How to use xG for smarter football betting
• Tennis maths: why winning 55% of points wins most matches

18+. Please gamble responsibly. For free, confidential support visit BeGambleAware.org.