There was no Wimbledon Championship this summer. Like many other top sporting events, the world’s oldest tennis tournament fell victim to the COVID-19 pandemic. While it may not satisfy purists, a Stanford University research team has responded with an AI-powered model capable of realistically simulating a Wimbledon final and more.
Inspired by the booming sports analytics trend which uses heaps of game videos to create predictive models for athlete behaviour, the Stanford researchers combined such modelling with image-based rendering to construct interactively controllable video “sprites” that mimic the styles and performance of top tennis pros.
“Our system can generate novel points between professional tennis players that resemble Wimbledon broadcasts, enabling new experiences such as the creation of matchups between players that have not competed in real life, or interactive control of players in the Wimbledon final,” the researchers explain in the paper detailing their approach.
Based on controllable video textures, the approach takes as its input a database of broadcast tennis videos that have been annotated with important match play events such as time and location of ball contact and type of stroke, etc.
The researchers focused on Wimbledon videos in building the database, which contains matches featuring popular players such as Roger Federer, Rafael Nadal and Novak Djokovic; as well as Serena Williams matches against Simona Halep and Camila Giorgi from the 2018 and 2019 Wimbledon tournaments.
There are several thousand shots from each player in the database. The researchers used these data points to model positioning and ball-strike decisions and build behaviour models reflecting how a given player positions themselves on the court and how and where they are most likely to hit the ball in a given situation.
Leveraging the database of annotated shot cycles, the researchers constructed statistical player behaviour models that input the point-state at the beginning of the shot cycle and produce player shot selection and recovery position decisions for the shot cycle. The behavioural models can select video clips that reflect actions the real-life player would be likely to perform in a given match play situation and generate video sprites that realistically depict a player’s appearance and style of motion and shot execution by capturing real-life strategies and tendencies.
The researchers say that according to expert tennis players surveyed, the rallies generated using their proposed approach are significantly more realistic in terms of player behaviour than video sprite methods that only consider the quality of motion transitions during video synthesis.
The researchers believe their system’s realistic game video generation capabilities and interactive user control features could be used to enable new experiences in sports entertainment and could also have practical applications in athlete visualization and coaching.
The paper Vid2Player: Controllable Video Sprites that Behave and Appear like Professional Tennis Players is on arXiv.
Reporter: Yuan Yuan | Editor: Michael Sarazen
This report offers a look at how China has leveraged artificial intelligence technologies in the battle against COVID-19. It is also available on Amazon Kindle. Along with this report, we also introduced a database covering additional 1428 artificial intelligence solutions from 12 pandemic scenarios.
Click here to find more reports from us.
We know you don’t want to miss any story. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.