When considering AI’s impact, the focus often gravitates towards job enhancement or daily life automation, yet its influence extends to less obvious areas like sports. Contrary to what one might assume, players aren’t the only ones making pivotal plays. A complex AI infrastructure plays a crucial role behind the scenes. Computer vision, a key AI technology, is transforming sports broadcasting. It provides enhanced analytics and real-time visual modifications that elevate the viewing experience significantly impacting how sports are consumed and enjoyed.

‍

What is Computer Vision?

Computer vision is a branch of Artificial Intelligence that allows computers to interpret information from images and video. The technology covers a broad range of disciplines, from object detection and image segmentation, to facial recognition and motion analysis. Applications of computer vision are vast, including areas such as autonomous vehicles, medical image analysis, surveillance systems, and augmented reality, each of which relies on the ability to accurately interpret and respond to visual inputs. One particular application of computer vision that has seen a booming public popularity for use is in the sports and broadcasting industry.

‍

Computer Vision in Sports

Whether it be Alan Turing’s imitation game in the 1940s or the advent of convolutional neural networks in the 1980s, Artificial intelligence and machine learning were long kept on the periphery of the public eye for those outside of a research setting. But as primarily a visual medium it was only a matter of time before Sports and Broadcasting took advantage of AI and Computer vision. After steady progress of research into its applications throughout the late 20th century, it wasn’t until 1997 that Computer Vision saw its first use in sports broadcasting with Prozone, a player tracking computer vision system used by Premier League Football club “Derby County”. This was followed shortly thereafter by Hawkeye in 2001, with their tennis ball tracking and line calling officiation system. The years following saw an explosion in the use of AI and computer vision to improve the viewing experience of sports. Following in the footsteps of major AI and machine learning breakthroughs of the period, computer vision techniques quickly became more advanced, and computation infrastructure they ran on became more affordable. Computer vision and AI in sport was rapidly becoming a must-have in any high intensity sport. And of all the use cases of CV in sports, positional tracking has arguably been one of the most influential applications.

How can a Computer See?

Much like the human stereoscopic vision, positional tracking with computer vision relies on multiple viewpoints of the subject. In order to “recognise” a subject, specially trained machine learning Computer Vision Models are used to detect the subject in each frame of video (whether that be a football, tennis ball, or a Human body). By recording the position of the subject in each frame, and accurately knowing the distance between them, 2 cameras can be used to provide X-axis, Y-axis, and depth information on a subject in the same way that human eyes perceive the world. 3 cameras spaced out around the subject then allow for accurate 3D spatial tracking in X,Y and Z axis, with additional cameras adding further accuracy to the system and higher frame rates allowing faster subjects to be tracked.

Consider a tennis course featuring two players. Surround the court with 10 cameras, each capturing 60 frames per second. Advanced software then processes the timing and positional data from each camera and every frame, focusing on the and recording the positions of players and the tennis balls. This data can then be combined and analysed to recreate a fully digital representation of the scene, allowing for extremely precise positional tracking of the tennis ball for use in Officiating and Line calling, or for overlaying interactive computer graphics or player metrics. Expound the same principle to any sport, and you can track Cars, Golf Balls, People, Rackets, Boats; and the versatility of the technology is clear.

‍

Computer Vision Hardware

At its heart, Computer vision in Sport requires a significant amount of compute Power. Earlier systems almost certainly required bulky and expensive on-site hardware to run their systems. The crux of the matter in computer vision is and always will be the scalability and cost of GPU (graphics processing Unit) compute power, as the accuracy of these systems is ultimately tied to two main factors; Frame-rate and image resolution. Both of which require increasingly more powerful systems as the technology advances, and therefore more money to invest and operate.

In and throughout the 90’s GPU were primarily developed and used for video games on PC and console systems. Slowly transitioning to general workload applications from 2000-2010. The use of high power GPUs have today become synonymous with AI and Machine learning. Additionally, cloud computing providers now offer various self contained GPU services for AI and machine learning, further bringing down the bar for entry by allowing for pay-as-you-go use of once extremely high investment cost GPU systems.

‍

Today in Sport

Today, we see widespread use of Computer vision across a wide range of different sports. From automated officiating and VAR (Video Assistant Referee), to advanced user metrics and positional tracking, AI has cemented its place in modern sports broadcasting as an essential service for viewers and Teams alike. Just a few examples include:

Tennis

With Hawkeye Leading the drive from 2001, Line calling in tennis has become a staple of accurate officiating in Tennis. Providing millimetre accuracy for line calling, transforming the way that umpires could officiate the sport, and provides a powerful visual aid for the audience.

‍

‍

Football

Although controversial among some groups of fans, VAR has been a major part of the officiating process in Premier and Championship leagues since the 2000s. Relying on a combination of Computer Vision tracking, and RFID Radio Frequency tracking, the accurate position of players and the ball have led to officiating decision accuracy increasing from 92.1% to 98.3% when compared to human referees.

‍

Advertising

A somewhat overlooked and “invisible” application of computer vision in sports, advertising boards can seamlessly have their advertising content adjusted depending on the region of the broadcast, without masking players or causing any kind of visual artifacting. It is a background application of computer vision in sports that while not immediately impressive to the audience, provides significant benefits to advertisers.

‍

Looking to the Future

Although Computer Vision has already revolutionised the way that sport is officiated, perhaps the most exciting prospect for sports fans are the many emergent applications of the technology. With ever more detailed and accurate player information, it is up to the imagination of established companies and Start-ups alike to utilise this data in new and exciting ways for fans.

For example, the positional player data provided by the National Football League in American Football has allowed the start-up StatusPRO to entirely re-create real-time games in Virtual reality, allowing fans to  experience the sport in a completely new light.

‍

Final THoughts

CV in sports is a proven hit in the industry, and its uptake and innovation does not look to be slowing down any time soon. With GPU compute resources becoming cheaper and more accessible than ever, it has never been a better time for businesses to explore the use of CV in any application. Once limited to the big players in sports, the accessibility of AI and CV technologies has already seen numerous start-ups cater to more and more niche sports and applications.

This is of course not only limited to sports. As the accessibility of the technology continues to improve, companies across all industries will be able to look to CV as an innovative solution to their problems. The potential for the technology is vast, and promises a future where its application is no longer limited by its cost, but by the imagination of those applying it.‍