AI on the Prize: Computer Vision and The Olympic Games

The 2020 Tokyo Olympic Games kicked off last Friday. As one of the most watched sporting events to occur worldwide, the Games will surely feature the use of the latest technology–not only by athletes and coaches but also by the broadcasting networks. We’ve written before about how computer vision is changing the way we understand sports from a statistical and strategic perspective–but how is vision AI changing the way we understand sports from our couches? 

The delayed 2020 Olympic Games will feature 33 different sports, 5 more than Rio and showcase 339 medaled events. One of the challenges with broadcasting such a wide array of competitions is creating visual cues for the spectator to help them understand the objective of each event. More likely than not when those visual tags appear, a vision AI and/or machine learning algorithm is at work. 

Models In Training

The computer vision algorithms we can expect to see at the Olympics, just like the athletes they are designed to analyze, required years of training. Instead of heading to the gym, computer vision algorithms rely on strong datasets to “learn” their objective. Datasets are created by taking a bunch of images and applying labels to help the computer tag or track when and where important events are happening. While the process of creating custom vision AI algorithms and models can be extremely complex (and piecemeal), industry leaders like Plainsight, have created intuitive vision AI platforms where dataset creation and model training can happen in the same interface–creating an optimized pipeline for real-world deployment.

AI For The Win

While computer vision and machine learning are hardly strangers to the world of sports, there are some exciting new advances that we can expect to make their debut at the Tokyo Summer Games: 

Gymnastics: The International Gymnastics Federation has announced that competition judges will use AI to help score athletes for the first time at the Tokyo Olympics. Developed by the Japanese IT company, Fujitsu, the AI will deliver a plethora of new performance data to judges in hopes that unbiased data will take some of the subjectivity out of the scoring process. Everything from height of jump, to body orientation during rotations, even the angle of the athlete’s arms on something like the rings will be available for judges to view. While the human judges will obviously have the final say in who the winner is, this new technology will definitely open up a new era in the sport of gymnastics.

On the Field: While a number of robots will be making their debut at the Tokyo games, expect to see the FSR, or Field Support Robot, on the track field. Developed by Toyota, the self-driving robots will help officials collect thrown objects like shot puts and javelins from the throwing field safely and quickly. Mascot Robots will also use computer vision to curate new spectator experiences, including social interactions and multilocation synchronized activities.

On the Track: The Track & Field viewer experience at this year’s games is getting a huge upgrade: Intel’s new 3DAT technology. Leveraging vision AI and multiple camera angles, 3DAT uses 3D pose estimation to create colorful visual overlays for athlete’s speed and other relevant race information–like who is winning. Fans will be able to analyze races like never before with new performance metrics like start speed and peak speed, just to name a few. This is going to be super cool, we can’t wait to see it in action.

Transforming Sports Before Our Very Eyes 

We can’t wait to see all this vision AI in action over the next couple of weeks! In addition to the use cases we mention above, we predict that viewers will see a lot more real-time velocity and trajectory recreations of anything whacked, thrown, hurled, or hit. Whether recreating the Gold Medal-clinching drive in Women’s Golf, or the World Record-shattering shot put throw, the technology to track the real-time movement of these objects is already being used in professional sports settings. And, we can expect to see it at the Games as well. We also wouldn’t be surprised if 3DAT visual tags show up in other arenas besides the Track, especially with the extra year of R&D that likely went into its development. 

It will also be interesting to see how the incorporation of vision AI technology with human judges pans out at the games. If applied as a decision-enhancing tool that combines the subjective human expertise of artistic judging with removal of the human factor from technically scored categories, vision AI can vastly improve the fairness of scoring for events. For instance if speed or power is something that is given a point value, the vision AI technology will be able to tell us exactly how fast or how powerful an individual athlete is. 

One thing is certain–Tokyo 2020 will definitely be a modern Olympic Games. Good luck to all the athletes and coaches, and Go Team USA!

More Plainsight Blog Posts:

Artificial Intelligence: Week #37 | 2021

Artificial Intelligence: Week #37 | 2021

This week in AI & Machine Learning: Robot waiters, Facebook new AI chip, data management for drug discovery, introducing TensorFlow similarity detection, PT-4 will have 100 trillion parameters, and more!

Artificial Intelligence: Week #36 | 2021

Artificial Intelligence: Week #36 | 2021

This week in AI & Machine Learning: The NFL tackles CTE with AI, A roomba that avoids dog poop, NOLS keyhole imaging, AI that writes its own code, deep reinforcement learning, and more!