ML-Powered Digital Referee for Basketball | Computer Vision Case Study

ML Video Processing for Community Sports: PlayHQ Case Study

NBA-like Game statistics and refereeing video app for basketball. Powered by ML.

Project Highlights

Innovative Solution: Developed an intelligent video processing system for community sports, transforming a single camera into a digital referee. This provided an enhanced viewing experience, mirroring professional sports production.

Player and Game Tracking: Implemented Convolutional Neural Network technology to track each player and the ball on the field. This capability directed camera focus and highlighted significant game moments.

Stabilized Image Quality: Created a specialized algorithm to mitigate camera shaking when zooming into crucial game moments, improving the overall video quality.

Industry: Sports

Location: Australia

Tech Stack:

TensorFlow
OpenCV

Offered Services: Computer Vision,
Machine Learning

Timeline: 3 months

Platforms: Web

See it live

CLIENT OVERVIEW

PlayHQ, Australia's fastest-growing sports tech company, serves as the preferred operating system for community sports nationwide, facilitating tens of thousands of games weekly. With over 1 million users and a reach spanning 60% of Australian community sports and 5500+ clubs and associations, PlayHQ is transforming the sports landscape. Their revolutionary approach replaces unreliable legacy systems with a seamless, mobile-first digital experience, all on one SaaS platform.

Notable sporting organizations such as the AFL, Cricket Australia, Netball Australia, New Zealand Cricket, and Hockey New Zealand are among the many who have joined the PlayHQ platform. PlayHQ's team of over 100 sports tech experts, based across Melbourne, Perth, Adelaide, India, and Ukraine, combines industry veterans from Australia's largest sporting codes and tech experts from successful start-ups.

Backed by experienced tech investors and significant sporting codes, PlayHQ is dedicated to continuous innovation and customer delight. Their mission is to harness the life-changing power of sports, bringing communities together, and making sport better, for players, parents, coaches, or administrators. This makes PlayHQ not just a company but a passionate team built for sport, by sport, and committed to the long game.

the challenge

The main idea was to create a digital referee. Community sports leagues have only one camera at their disposal to film the game, unlike professional sports.

‍

We had to teach the camera to follow the game and to be able to zoom in on the important moments so that we could get content that was fun to watch, as if the game had been filmed with professional production.

core solution

Detect each player separately

Using Convolutional Neural Network, we trained the system to distinguish between players on the field so that it could help us point the camera in the right direction.

It is essential to watch the ball

We started detecting the ball separately so that the camera could better understand which way to look and which moments of the game needed to be highlighted.

Understand the weight point of interest

There are quite a few players on the field. And we also had to teach the system to understand where to look.

Image quality

In the process of work, we faced the fact that zooming in on important moments of the game, the camera shook. To solve this problem, we wrote a special algorithm, which later helped to get a smoother, better picture.

Product Improvement

Upon the successful implementation of the fundamental functionality, as initially planned, we embarked on a journey to enhance the system. The objective was to enable the system to collect exhaustive game statistics.

The goal was to engineer a system akin to the sophisticated data collection mechanisms employed by the NBA. However, it was here that we encountered the crux of the challenge. Unlike the NBA, equipped with myriad cameras and sensors positioned strategically, we were limited to a single camera.

This scarcity of data collection means drastically curtailed our ability to develop the envisioned functionality. Ultimately, due to the impracticability of continuing with insufficient resources, the project was paused, while the solution proved its efficiency and applicability for basketball and other sports.

The goal was never to expend exorbitant resources but rather to create a highly efficient system within the constraints of a limited budget and minimal data.

The outcome

The Uinno team made substantial strides with this experimental project, extracting maximum potential from available resources. Our innovative solution significantly enhanced the content quality for community sports across Australia, successfully meeting all initial objectives. In our quest to overdeliver, we aimed to construct a system resembling the NBA's data acquisition.

However, the data paucity proved a hurdle too significant to overcome. Nonetheless, given adequate data, we were well-prepared to accomplish this ambitious task. Although the project was temporarily shelved at the client's request, our collaboration continues, tackling fresh ideas and challenges together.