We built a Computer Vision Pipeline to automate video data extraction for an Ice Hockey Analytics firm.
We worked with an ice hockey analytics firm specializing in data-driven insights and advanced statistical analysis. They are dedicated to improving team performance, player evaluation, and strategic decision-making in professional ice hockey organizations.
.webp)
As a team sport, ice hockey relies on a collaborative dynamic. Goals, faults and overall performance are the result of team effort and individual dedication.
With that in mind, the client needed a project plan to calculate each player’s amount of time playing on the field, defined as Time On Ice.
For this stage, our strategy primarily involved leveraging existing data sources to compile a comprehensive dataset. As customer is already operating and manually annotating videos, past games were a vast source of available data.
Additionally, in some cases, we needed to supplement the dataset by labeling additional information. This meant identifying gaps in the current information and proceeding to add more labels to enrich the dataset's depth and quality. By employing this dual approach, we were able to improve the dataset precision for the project's needs.
We developed and implemented a comprehensive multi-stage process that integrated advanced video analytics and Computer Vision techniques. This approach was tailored to detect players, identify their jersey numbers, track positions on the field, and predict Time On Ice (TOI) for each individual player.
Through this systematic process, we achieved notable advancements in player tracking, number recognition, and positional monitoring, facilitating enhanced insights and analysis in sports performance evaluation.
Over a span of nearly three years dedicated to the project, our team engaged in more than 70 development sprints. Impressively, approximately 80% of these sprints demonstrated significant and noteworthy improvements.
We were able to optimize system speed and performance. By tracking only game-play segments, synchronizing tasks, and streamlining computations, we cut processing time and cloud computing cost significantly.
The developed pipeline helped the client to collect data from sport videos in an automated way, saving time and resources. Remarkably, it allowed the client to secure a collaboration contract with a media platform, and scale up the number of games analyzed each week.
Now, we aim to scale up the model working with next iterations of the project, making it faster, more precise and able to extract additional information.
“With eidos.ai’s expertise in Computer Vision and sports, we have been able to achieve remarkable results, made significant progress and solved challenges with a collaborative approach”
