Simulation of interaction patterns and simulative effectiveness analysis of creative plays in team sports by means of neural networks

• Applicant: Professor Dr. Daniel Memmert
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2008 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 66106926

The characterization and analysis of complex games like football cannot be reduced to static distributions of numbers but requires the consideration of the game´s very own dynamic processes. Therefore, success of specific actions cannot simply be stated in percentages without giving the context in which these actions have taken place. Such analysis tools are not yet available. In the course of the underlying project, concepts of earlier projects shall be used as a basis to develop a computer-based analysis procedure that recognizes process patterns in a game and evaluates their effectiveness. The particular focus is on creative processes that do not occur frequently enough to be recognized, neither by a qualitative-context nor by a quantitative-statistical analysis. In order to analyze the effectiveness of such rare process patterns, the development of a specific game-simulation tool is planned: this way, the impact of creative actions on the further course of the game can be simulated and analyzed, both online-manually and offline-systematically. The project outline can be divided into two core areas:* The simulation serves the generation of action processes and their effectiveness analysis based on data collected from real matches.* The creativity analysis uses the results of the simulation specifically for the recognition, generation and evaluation of creative action processes and poses the basis for training of creative actions in the match practice.In preparation of the project, software tools have been developed that will need to be adjusted and developed further: The analysis tool SOCCER combines conventional data analysis, dynamic modeling based on states and events and artificial neural networks. Here, events such as win or loss of the ball are calculated from player and ball positions. From these positions, neural networks can also detect time-dependent player positions, thus enable the embedding of individual movements and its superordinate processes (e.g. defensive teamwork, winning the ball back, or attacking) into the respective situational context. By means of appropriate success-indicators such as ball possession, actions and processes can automatically be evaluated with regard to success and both quantitative as well as qualitative analyses will be possible in situation-specific contexts. More difficult is the analysis of creative actions and processes which are mostly characterized by their unexpected and rare occurrence and thus are difficult to identify within the dynamics of the game when using conventional recognition methods. For this, TriTop (Triangular Topology), a new type of a self-organizing network has been developed. With the help of a fractal triangular topology and a dynamic generation of neurons, the program is able to distinguish between rare and frequent events without losing the cluster landscape of neurons that characterizes the process-distribution.

• DFG Programme: Research Grants
• Participating Person: Professor Dr. Jürgen Perl