Sporting success thanks to AI

What positions do the players take when in possession of the ball and what other phases of the game usually result from this? When and where is the opponent put under pressure? What system of play, including running routes, has the team practiced? Which moves lead to goals? Data is analyzed to answer these questions. In professional soccer and handball, more and more data has to be processed and interpreted in less and less time in scouting and match analysis. Artificial intelligence methods have therefore become an integral part of sport and are becoming increasingly important. Based on video, event and position data, scientists have developed and validated various tools that can be used to extract and evaluate sports match data completely automatically, extremely quickly and reliably.

Technological advancements in sensor technology and (semi-)automated data collection allow for comprehensive analyses in sports such as soccer, basketball, and handball. Modern tracking systems are used for load control in both competition and training. The Institute for Exercise Training and Sport Informatics at the German Sport University Cologne (DSHS) and the L3S research center at Leibniz Universität Hannover have now successfully completed the project "MM4SPA - Multimodal Analysis for Sports Analytics: Intelligent Synchronization and Semantic Enrichment of Position and Video Data for the Analysis of Sports Game Data", funded by the Federal Ministry of Education and Research (BMBF).

The aim of the project was to develop machine learning methods for the multimodal semantic analysis of sports games from different domains in order to provide experts with high-quality analyses. The project was positively assessed as part of the "Application of Artificial Intelligence Methods in Practice" funding line and funded with more than 800,000 euros over a project period of two years (2020-2023). With this funding line, the BMBF supports projects that aim to transfer innovative research results from the field of artificial intelligence into practice.

"In particular, we were able to further develop the exact synchronization of event, video and position data, which is essential for the development and applicability of multimodal approaches. In addition to more granular cross-sport definitions and solutions for event recognition, we have implemented an open science framework for processing sports data," says Prof. Dr. Daniel Memmert, Managing Director of the Institute of Exercise Training and Sport Informatics at the German Sport University Cologne. He and his team have been working for years on the analysis (e.g. with SOCCER©) and simulation of complex position data (so-called big data) and have developed and validated theoretical framework models for this purpose (cf. DFG project "e-Science").

Another project partner is the Munich-based technology company KINEXON, an expert in the field of real-time collection and processing of sports performance data, with whom the sport university has been working for five years. As part of this cooperation, the Institute of Exercise Training and Sport Informatics is one of the few research facilities in the world where position data can be recorded in indoor and outdoor sports. The current project has succeeded in integrating machine learning methods into KINEXON's existing analysis platform. "The methods developed were integrated into our software as a prototype in order to make the analyses available to the sport and our partners," says Dr. Maximilian Schmidt, Managing Director of KINEXON Sports & Media.

The necessary expertise in handling video data and machine learning techniques for the project was provided by the L3S Research Center, located at Leibniz University Hannover. Here, computer scientist Professor Dr. Ralph Ewerth, together with his research group 'Visual Analytics,' has been involved for many years in the analysis of videos and data presented in multiple modalities (e.g., language and image) or captured by different sensors. Among other things, the working group researches and develops methods for automatic video analysis and artificial vision using machine learning processes. "By using precise pitch registration as the basis for the semantic linking of video and position data, we were able to advance methods for recognizing phases of play, such as pressing, for further match analysis," says Prof. Ewerth, outlining the Hanover team's field of application.

Project contact:
Univ.-Prof. Dr. Daniel Memmert
Dr. Robert Rein
Institute für Exercise Training and Sport Informatics, DSHS
Tel.:  +49 221 4982-4330
E-Mail: memmert@­dshs-koeln.de

Prof. Dr. Ralph Ewerth
L3S Research Center, Visual Analytics working group,
Leibniz University Hannover
Tel.:  +49 511 762 19651
E-Mail: ewerth@­l3s.de

Dr. Maximilian Schmidt
KINEXON Sports & Media GmbH
Tel.: +49 89 200 61650
E-Mail: maximilian.schmidt@­kinexon.com