Evolution and Application of Groups of Trajectory Sets

Abstract The work presented in this paper continues a previous approach to automatically detect tactics based on spatio-temporal data in the context of team handball. It will be shown how the availability of more data allows to verify the principal approach. However, it will also be shown, that the previous approach for choosing parameters of the applied methods was suboptimal and an application-oriented approach based on heuristics helps to improve the results significantly. Like in case of the previous publication, it is important to realize that the combination of two methods is used as a substitute for a classification approach which would usually be used in the given application scenario. However, there is currently no means to generate the necessary training data which results in the need to apply descriptive methods like clustering and the search for frequent itemsets. The latter is needed because there is no sensible notion of distance for sets of trajectories.