AISportsWatch: Livestreaming sports at scale with Google Cloud

About AISportsWatch

AISportsWatch offers a cost efficient and fully automatic production, activation, and broadcasting service that delivers video content to fans of more than 2,000 sports clubs.

Industries: Technology
Location: Germany

About Go Reply

Part of the Reply Group, Go Reply specializes in the development and implementation of cloud platform architectures and the creation of solutions based on Google Cloud.

AISportsWatch uses Cloud Functions and Compute Engine virtual machines to power an all-inclusive, AI-led sports broadcasting platform for fans of all kinds of sports.

Google Cloud results

  • Orchestrates a services-based architecture using Cloud Functions
  • Automates livestreams using AI technology powered with Compute Engine virtual machines
  • Uses the latest graphical hardware to capture more data and statistics for fans, players, and clubs

Livestreams 300 matches every weekend in full HD

Watching a favorite sports team is one of the few truly global pastimes, with every society having some kind of sports culture. Top clubs can generate revenues in the hundreds of millions of dollars and broadcast every match to fans across the world, but the matches that are broadcast on TV or mainstream platforms represent only a tiny fraction of the sport that is played. AISportsWatch aims to balance the scales by providing an online platform for broadcasting the matches of amateur and smaller teams across a range of sports. At this level, club resources might not stretch to handling a video production unit, but their fans are no less passionate and no less deserving than the fans of the bigger teams. AISportsWatch developed its own cutting-edge cameras that provide an unbroken 180-degree view of a playing field. Alongside the hardware, AISportsWatch worked on a range of AI tools that could automatically create an edited live feed, without the need for expensive equipment or a team of producers. The result is a streamlined, affordable, one-stop sports broadcasting platform.

“The challenge for us is to maintain a consistent level of quality across all our livestreams. As we grew bigger, we started to have problems with stability and outages. We needed a new infrastructure that could scale on demand without compromising our standards. For us, that was Google Cloud.”

Marvin Baudewig, CTO, AISportsWatch

Launched in 2017, AISportsWatch began streaming football matches in its native Germany and has grown quickly since then. As well as broadcasting games, the company’s platform also collects data on everything that happens on the playing field, helping clubs to refine their coaching strategies and pinpoint areas for player training and improvement to focus on. The platform also comes with tools that allow fans, players, and teams to share their favorite clips on the social media platform of their choice. Today the company partners with more than 2,000 clubs to livestream football (soccer), basketball, and, most recently, ice hockey matches in full 1080p HD. To achieve such an all-encompassing platform at such scale, AISportsWatch turned to Google Cloud.

“The challenge for us is to maintain a consistent level of quality across all our livestreams. As we grew bigger, we started to have problems with stability and outages,” says Marvin Baudewig, CTO at AISportsWatch. “We needed a new infrastructure that could scale on demand without compromising our standards. For us, that was Google Cloud.”

Unique solutions for unique challenges with Google Cloud

What makes AISportsWatch so valuable is its simplicity. If a club wants to record and broadcast its matches, traditionally it would have to invest in multiple cameras, an editing bay, and hiring a production team. But with AISportsWatch, it can simply hook up a single camera and sign up to the online platform. The cameras are built with multiple lenses, arranged side by side, which together provide full 180-degree coverage of a sports field at all times. The company then uses AI to automatically knit together the images from the lenses, identify where the main activity is, and focus the video feed there. The result is a seamless, edited, professional-looking live feed that ensures viewers never miss the action.

When AISportsWatch first launched, it covered a few dozen matches every weekend. Even then, an on-premises infrastructure didn’t have the capacity to broadcast livestreams without investing a huge amount into hardware, so the company built its initial platform on the cloud. By mid-2018, however, AISportsWatch increased not only the number of matches it was broadcasting but also the type of sports it was covering. At the same time it was working on new, higher quality cameras and improved AI that ingested and analyzed even more data, which needed even more resources. The existing cloud infrastructure started to prove unreliable, with unplanned outages coming at crucial times when demand was at its highest. In one of these cases, AISportsWatch had to resort to manual methods in order to deliver an unexpected 50,000 parallel accesses from football fans tuning in to watch their favorite team.

“Cloud Functions makes it easy to deploy code, so it’s a real highlight for us. It’s fast, cost efficient, and we know it can scale up to cope with high request volume, without the need for us to get involved. It’s made our lives much easier.”

Marvin Baudewig, CTO, AISportsWatch

AISportsWatch looked for an alternative cloud platform and chose to trial some of its services with Google Cloud. “We were up and running very quickly with Google Cloud,” says Marvin. “We found it worked really fast and well for us.” After a few months, AISportsWatch was ready to complete the migration, teaming up with Go Reply, a Google Cloud Premier Partner, to help reshape the company’s architecture for a smooth switchover to the Google Cloud environment. Adapting quickly, AISportsWatch completed its migration within a month.

The company took advantage of the flexibility of Google Cloud to design an infrastructure that can adapt to the unique challenges of its work. For the livestreaming, the vast majority of the traffic is handled with microservices orchestrated with Cloud Functions, while the stateful applications run on App Engine.

“Cloud Functions makes it easy to deploy code, so it’s a real highlight for us,” says Marvin. “It’s fast, cost efficient, and we know it can scale up to cope with high request volume, without the need for us to get involved. It’s made our lives much easier.”

The AI tools run on Compute Engine virtual machines, which can scale up or down as required. Meanwhile, Cloud Storage provides a secure, easily accessible way of storing the huge amount of streams that AISportsWatch has accumulated over the years.

Scaling up with smoother service

With Google Cloud, AISportsWatch has expanded its business in all directions and improved its service to customers. Today, it livestreams around 300 matches every weekend. Since moving to Google Cloud, the company has accumulated around 220 terabytes of data including video and match statistics. Even as the load on its infrastructure has increased, AISportsWatch has been able to make its livestreams an even smoother experience for customers. “With Memcache in App Engine, we’ve reduced our page load times by 40% and kept them constant, even with high traffic loads when we’re streaming 50 matches at the same time,” says Marvin.

Perhaps the most crucial difference in terms of performance is the stability of the new platform. “We used to crash for maybe an hour every week, but since we moved to Google Cloud, the infrastructure hasn’t failed at all,” he says.

“Not only can we analyze and track the game in real time, we can also determine the players and ball position, while simultaneously providing the user with a full HD stream. None of it would be possible without the flexible scalability of the GPUs available with Google Cloud.”

Marvin Baudewig, CTO, AISportsWatch

Google Cloud has a simplicity and ease of use that has improved life for Marvin’s team of developers. The combination of Cloud Functions and Compute Engine has cut deployment times from 30 minutes to around five minutes, leading to more updates and features for AISportsWatch’s customers to enjoy. One popular example is a coaching tool that can automatically generate highlight clips and game summaries in real time.

Also, the ability to customize the hardware on its virtual machines on Compute Engine allows AISportsWatch to ingest and analyze even more information than before. “Not only can we analyze and track the game in real time, we can also determine the players and ball position, while simultaneously providing the user with a full HD stream,” says Marvin. “None of it would be possible without the flexible scalability of the GPUs available with Google Cloud.”

As an energetic start-up, AISportsWatch is constantly working to improve itself and create new experiences for its users and partners. Marvin has started to experiment with Cloud Run for resource-intensive tasks. The first of these is a tool that will automatically take video clips of a match and turn them into GIFs, making it easier for fans and clubs to share the most exciting moments on social media.

AISportsWatch is also looking beyond its proprietary AI tools and experimenting with Video Intelligence API which could help the platform track players and objects more precisely, for more useful statistics. Growth is still the main objective for the company, which plans to stream around 1,500 matches every weekend by the end of next year, a fivefold expansion. “If we want to keep growing, it makes sense to keep using Google Cloud,” says Marvin.

About AISportsWatch

AISportsWatch offers a cost efficient and fully automatic production, activation, and broadcasting service that delivers video content to fans of more than 2,000 sports clubs.

Industries: Technology
Location: Germany

About Go Reply

Part of the Reply Group, Go Reply specializes in the development and implementation of cloud platform architectures and the creation of solutions based on Google Cloud.