Beginning in the early 1990s a number of tracking systems for team sports have been developed (Santiago et al., 2010). To record the player trajectories, these systems usually rely either on video footage from the games, which is processed using software tools, or on sensors which are attached to the players. The collected data needs to be post-processed to be usable by e.g. players, coaches, or others interested in team or player performance. In this work we will use data recorded by the Sports Performance Analyzer`s (SPA) tracking system (Wilhelm et al., 2010) and illustrate its methods for the performance analysis, varying from low level statistical trajectory analysis to higher level play recognition and matching. The SPA tracking system consists of two overhead cameras in the sports hall which record the game from a bird`s eye view point. The video streams are used to track the players offline, and the players` trajectories are available for further analysis (see Wilhelm et al. (2010) for a comprehensive introduction). In this contribution, first we explain a method for the automatic segmentation of a game into action and break phases using the recorded trajectories. Based on this data we can compute the players` covered distance and running time for the net game time (i.e. action phases when the game was actually running) and for the complete game including breaks. This data can be further divided into speed classes for a more detailed analysis of the running efforts, and broken down to single players, parts or teams. We are also able to detect so-called run-phases, where one team scored several points while the other team did not score at all, and restrict the aforementioned analysis to these phases. Our second analysis technique is the extraction of specific plays from the trajectory data. These plays are stored in a database and similar plays are filtered out using self organizing maps. We applied the play extraction and matching algorithms to recorded basketball games from the German major league and were able to extract and match single plays. In 18 analysed games from three different seasons, the automatic net-time segmentation was on average 90.0% accurate. The distance and time analysis gives a fast way to extract valuable and easily interpretable information from the trajectory data. The information about running distance and time gives insight into the physical demand on the players in a specific sport. By analysing a series of games we are able to generate a detailed activity profile for the individual players, which could be used to fine tune their training. Being able to extract and match similar plays gives us the possibility to evaluate the performance of a specific play by e.g. computing the average number of points scored in this play.
© Copyright 2012 World Congress of Performance Analysis of Sport IX. Veröffentlicht von University of Worcester. Alle Rechte vorbehalten.