I will be talking about using spatio-temporal information for the analysis of tennis matches. State-of-the-art tennis modelling techniques use player statistics such as the percentage of points won in the first serve or double faults. Advances in computer vision techniques and technology have made it possible to obtain spatio-temporal data from tennis matches. It is now possible to get the 3D position of both players and the ball, the shot speed and angle and many more characteristics. In my work, I use this data in order to build better models. For instance, one of the key differences with the state-of-the-art tennis predicting models is that we can now look at intra-point dynamics.
In this talk I will first describe the state-of-the-art approach to modelling tennis matches. Then I will give an overview of some computer vision and machine learning techniques that can be used to extract spatio-temporal data from tennis videos. Finally I will talk about how these can be used to build better tennis models.