We offer an 'Optical Tracking' service that provides user with positional and physical data that is captured through video.
Simply put, this means that artificial intelligence(AI) is used to recognise players, track their movements, and turn this into data.

‍

‍

For tracking data collection, we use computer vision to turn video frames into positional data using an analytical model. This requires over 100,000 frames to be analysed for each match, from which over three million data points are collected.

To do this analysis we've built models using machine learning, by which a model can improve through experience. By providing a larger amount of high-quality annotated data, the model can 'learn' and complete its task more efficiently. For us, this means our model being able to better recognise the individual players as the move around and on and off the pitch.

‍

‍

 Recently, we’ve decided to adopt MLOps to support our tracking data project. This is an approach to Machine Learning that brings together the development and operational sides involved in developing an effective model. This covers many areas, including the model’s code, data collection, automation and more.

We have been able to reduce the length of time required for tracking data to be collected for a match thanks to taking an MLOps approach.It has helped us reduce the number of errors and therefore the amount of time needed for manual corrections to be made.

‍

‍

We’ve been impressed with how adopting it had supported us so far. We expect MLOps to continue to improve the efficiency of our tracking data collection process and we will look to apply it to more projects in future!

‍

Read more about how MLOps has helped us in our Medium article here.

‍