Computer Vision for a Competitive Edge in Sports

SUMMARY

• Our client develops cutting-edge technology to assess skill and provide insights to athletes and decision makers in the sports industry.

• Hop was engaged to solve the complex and layered problem of limb-tracking – pioneering research work in the realm of computer vision.

• This collaboration played a key role in the research and development of the 3D aspect of our client’s technology, currently in use by a number of professional sports teams to improve performance. 

THE COMPANY

In professional sports, a large part of success is tracking performance – executives, coaches, trainers, and players all have an interest in assessing skills and making improvements. To do this well, a lot is required in terms of resources, and has been manageable for only the top teams and athletes. Over a number of years, Hop partnered with a sports and entertainment company innovating technology to measure the nuances of skill, providing insights for better decision-making at all levels of an organization.

THE CHALLENGE

In our R&D work with this company, we focused on using machine learning for high-value analysis of performance in a semi-automated manner. Our primary challenge was to apply computer vision to track every limb of every player in 3D. This is the foundation for learning patterns of movement over time, of both individuals and teams, that inform decisions of coaching and training staff.

The challenge of tracking every limb of every player is actually many computer vision problems rolled into one – 2D and 3D – making this a complex problem that’s hard to solve in a reliable and robust manner. In particular, athletes move in ways that prior research in computer vision has not yet addressed. We therefore had foundational problems to solve that then surfaced other new problems, creating layers of complexity that each became open areas of work.

As these layers emerged, so did a pattern inherent to production-scale ML that had to be addressed in parallel – with nothing ever being “done”, open areas of work needed to be revisited as progress was made, and procedures had to be put in place along the way for this purpose. We also grappled throughout the project with the translation of academic or idealized assumptions to real-world contexts, where, for example, a camera may have optical aberrations, or be so heavy that its weight can affect its positioning over time.

THE APPROACH

To tackle these challenges, we assembled a cross-functional and nimble team, with skills in R&D, engineering, and operations. Our workflow was defined by rapid iterative loops with real-world data and frequent demos to stakeholders, through which we were able to learn quickly and iterate progressively towards functioning systems. Throughout the years we contributed to this project, we developed rigorous quantitative measures to guide our work, keeping our team on track as we navigated unexplored territory.

Response from the CEO to a demo of our work:

THE RESULTS

While this client had their own technical team, they were able to extend their research capacity and accelerate their timeline by working with Hop. It was remarkably rewarding for our team to work on this industrial-scale computer vision project and see how our progress deepened the problem. Hop’s research and development of the 3D aspect of this technology played a key role in bringing it to the real world — it’s currently in use by a number of professional sports teams to improve performance. 

Is your team looking to extend its ML research capacity? Contact us to learn how Hop can help.