Sports Analytics: Hockey Use Case

Discover how we helped Reap Analytics by accurately annotating players, rink markings, jersey numbers, and pucks to optimize the training of their AI models.

Before After after annotation

For years, generating sports statistics on player performance was a manual process.

Now, with the power of Computer Vision applied to simple game footage, Reap Analytics is set to revolutionize the way game stats are produced.

Their goal? To streamline the NHL draft process by providing more accurate and efficient insights into player performance.

Context

CONTEXT

Our client, Reap Analytics, provides NHL teams with reliable player statistics.

description of puck and player position

Screenshot taken from Reap Analytics marketing video, that is showing the precise description of player and puck position.

Reap Analytics has developed advanced algorithms to track player positions and puck movement, turning this data into precise coordinates that can be used for analysis. 

However, to train these algorithms effectively, reliable data is essential—especially clear identification of the puck’s position and the player numbers on their jerseys and sleeves.

The annotation process itself involves detecting players and rink markings, identifying and translating player numbers (on both the back and sleeves), and tracking the puck, particularly during key moments like shots and goals.

To scale up their data annotation efforts, Reap Analytics tested multiple annotation providers before choosing People for AI as their trusted partner.

OUR SOLUTION

OUR SOLUTION

To annotate the data, we built two expert teams: one focused on tracking player numbers, and the other on detecting pucks. This approach allowed us to efficiently handle different annotation needs at the same time.

We also established a robust annotation process, from carefully selecting key game sequences (like shots and goals), to tracking pucks, detecting player numbers, and even developing an error detection algorithm to simplify corrections directly within the annotation tool.

Thousands of video clips processed!

Two expert teams with 15 trained annotators specializing in this field.

Our methodology has proven effective, starting with detailed annotation guides and refining them based on the questions that arose during the process.

Matthieu Warnier

Data Labeling Director @ People for AI

« The vast amount of data to be annotated posed a major challenge, which we successfully overcame by dividing the annotation process into multiple stages and pre-selecting the sequences to be annotated. This approach ensured both the efficiency and accuracy of our teams. »

OUR IMPACT

OUR IMPACT

The number of game sequences annotated increased by a factor of 100 compared to in-house annotation.

Improved quality thanks to the development of an algorithm for detecting player number errors.

Reap Analytics was able to refocus on higher value-added tasks.

We have led this project from the start by:

Developing annotation guides to support the accurate labeling. Covering methodology, special cases, and the management of certainty and uncertainty.

Creating an automated tool for detecting number errors, based on player coordinates and specific functional rules.

By outsourcing data annotation to a specialized company, Reap Analytics was able to focus on promoting and positioning its tool to the top NHL teams.

Joe Gratz

CEO @ Reap Analytics Client of People For AI since 2023

« After testing several annotation companies, I naturally chose People For AI, which stood out for its superior annotation quality. No other company matched PFAI’s expertise in both annotation and understanding of hockey. I also appreciated the immediate trust established with its Director, Matthieu Warnier. »

Our ability to train two expert teams for complex annotation tasks enabled Reap Analytics to develop a robust AI model, delivering reliability beyond expectations.

Our labeled data will exceed your expectations.

They trust us

We provide training data. We do not provide your data to anyone else and we will not subscribe you to any newsletter without your consent.