News-Service - Newsletter abonnieren - STEMMER IMAGING

Keeping you one step ahead


Subscribe to our updates >

Read more

Why are vision systems so interesting for sports applications?

15 Mar 2023 | Reading time: 3 min

When athletes compete against each other, vision systems are often involved to check referee decisions, create statistics, or make training more effective. Because with high-resolution images, details can be recognised that are hardly visible to the human eye: Whether it is the smallest changes in body posture, the exact trajectory of a ball or minimal distances on the goal line – computer vision makes sport fairer and more exciting.

Maurice Lingenfelder, Team Leader Project Management, explains why and which vision solutions are so interesting for sports and how the STEMMER IMAGING project team can support customers from the sports industry.

Maurice Lingenfelder, Team Leader Project Management, STEMMER IMAGING

STEMMER IMAGING has seen a huge increase in customer projects in non-industrial environments, especially in sports, in recent years. Why are vision systems so interesting for sports applications in particular?

Digitisation and the accompanying visibility and shareability of data in everyday life is becoming increasingly normal for people, especially in sports. Very few people go for a jog these days without at least recording distance, time, or pulse. And for many users, it has become a matter of course to share such statistics about their own performance with their personal environment. At the same time, there is a constant urge for newer and better statistics on the performance and current level of athletes in all sports.

Computer vision plays a major role in satisfying this demand for data. It provides inexpensive ways to automatically and reliably obtain information that would previously have required entire television broadcast teams or sports physicians.

In addition to the competitive aspects of sport, however, there is increasingly a link between sport and entertainment. Sport is being gamified through live tracking and augmented reality in special indoor facilities.

The target group thus goes far beyond the athlete and his personal performance; the social event is in the foreground. And here, too, recorded data are indispensable as the basis for any evaluation. Such facilities must therefore be equipped with vision sensors and suitable software.

You just said that computer vision provides inexpensive ways to get information automatically and reliably. Can you explain this in more detail? What are the possibilities, for example?

A typical task in sports applications is tracking the position of athletes or objects (e.g. balls). It really doesn't take much. In the simplest case, a single standard 2D area scan camera with lens, cable and PC is sufficient to determine the position.

And the quality and certainty of the positions found increases rapidly with the addition of a few more cameras. Then powerful software is needed. And with the increasing availability of suitable training sets or neural networks, this too is no longer as time-consuming and cost intensive as it used to be.

It is important to mention here, however, that even the best artificial intelligence can only evaluate what data is actually available in the images. So there are definitely limits to the usefulness of the component prices that one should be aware of. In addition, the reliability of the data is essential in many applications.

Just imagine the scandal of transmission errors of the cameras in the World Cup final causing the goal line technology to fail at the decisive moment. In series production at the latest, it is therefore worthwhile to rely on standards that have been tried and tested for years. And that is what machine vision can offer.

Does the approach to a customer project in the non-industrial environment differ from that in the industrial environment?

In contrast to the industrial environment, which has already been using machine vision applications on a large scale for many years, machine vision in the non-industrial environment is still in its infancy. The concrete application and the starting point for a joint co-development must first be identified. Sometimes there is only a basic vision in mind; sometimes there are already functional prototypes that are then robustly developed for series use with our help. Therefore, it is even more important than ever to take the time at the beginning of a project to clearly identify the necessary requirements with the customers and to define a common goal that can be realised under the general conditions.

It is often smaller SW or AI companies that venture into such new topics. They are particularly dependent on broad support throughout the entire course of a development project, as they lack relevant experience or networks in machine vision.

A clear trend is therefore that the tasks are clearly distributed in a co-development. Our customers can rely on us to provide them with a complete image acquisition subsystem from a single source, and they can focus entirely on their core competence of image evaluation, data acquisition and display.

When finding technical solutions for the sports sector, we also must take into account the fact that applications are almost always associated with high volumes; therefore, the price of the final product carries even more weight. The compactness of the final subsystems also plays a much greater role in these applications. Here, it helps us a lot that we recently launched an in-house developed modular embedded board that can be used in many sports applications and combines the required compact design with sufficient computing power for AI applications.

Can you describe the Modular Embedded Board in more detail?

The board is based on the powerful NVIDIA Jetson hardware and offers plug-and-play efficiency for the rapid implementation of embedded vision and AI projects. It is equipped with a Tensor Core GPU that enables massive acceleration.

It is fully GenICam compatible, which means a comprehensive choice of cameras for all common interfaces such as GigE, USB or MIPI without special programming knowledge.

The integrated TCP offload technology offloads the processor while enabling maximum performance with low power consumption. The board is part of our comprehensive STEMMER IMAGING Modular Embedded ecosystem, which combines a powerful hardware selection with the best software tools and individual service packages.

Our project team's services include requirements analysis, solution design, feasibility studies, SW or HW development or production planning. In this way, we distinguish ourselves internally from classic component distribution, where only individual additional services are offered.

Can you give us an example from the sports sector that you have implemented in your team?

Together with a customer, we developed a compact sensor unit for football stadiums that sends processed, high-quality video recordings from several cameras via a cloud to a host computer. This evaluates the recordings with the help of artificial intelligence to automatically generate player statistics and make them available to TV broadcasters, among others.

Our task was to design a system layout that included an optimal selection, number, and orientation of vision components to cover the desired requirements. Image acquisition from each camera had to be synchronised and camera and lens settings such as exposure time and focus had to be automatically adjusted according to weather and daylight. A communication interface for initial setup of the cameras via remote was desired, as the unit was mounted on the roof where access is difficult.

The camera unit itself needed an integrated processor module so that the recordings could be evaluated, processed, and compressed before transfer to the host to ensure real-time transmission. This is where our embedded board delivered great benefits. The design and manufacture of a suitable housing was carried out by one of our partners while STEMMER IMAGING offered the procurement of all components, as well as assembly, configuration, testing and labelling of the unit reducing many of the challenges faced by the customer.

Last but not least, a look into the future: Where do you think the journey will lead with regard to vision systems in the sports environment?

I am convinced that the combination of inexpensive and compact sensor technology and constantly improving AI algorithms will lead to what’s currently available in professional sport being made viable for amateur athletes in a corresponding form.

Cloud video transmissions and live statistics from amateur sport will mean that interested grandparents can easily follow their grandchild's district league match live from 1000 km away. I will be able to watch the tennis match with my club colleague enlarged and in slow motion directly afterwards on my mobile phone.

Or I will be able to share an interactive 360° image of my last skiing holiday from the fun park with my friends. All these lifestyle topics can be made accessible to everyone at acceptable prices through computer vision.

At the same time, professional sports will continue to drive innovation and demand more and more data to gain competitive advantage. I am sure we can expect many more surprising applications and innovations. And we look forward to doing our part to turn such ideas into reality via collaborative development projects.


Big Data and Artificial Intelligence applied to Sport

Artificial intelligence and big data are two of the most disruptive technology innovation of our lifetime, and there is a lot of discussion going on about their potential and possible applications. Find out how these trends have revolutionized sport and what role computer vision plays in this.

Download >