FWF - MASA - Movement and action sequence analysis in complex sport games

Due to the combination of PCs and digital video the systematic analysis of sport games has largely improved during the recent years. Analysis systems currently used in science and sport practise can be divided in two subgroups. First, calibrated multi camera systems provide information about the physical load of athletes during competition and training. The results help to estimate the physical demands of athletes (e.g. number of sprints or jumps). Second, different technologies are used to record single events during games. The results are performance parameters like goals/shots. Although such attempts of game analysis are already very useful for scientists and practitioner they still have noticeable shortcomings. Systems which allow to record movement patterns with a sufficient accuracy are related to an enormous amount of technical and personal demand. Multi camera systems have to be connected via hardware and calibrated. Furthermore, automated tracking algorithms are vulnerable to errors due to occlusion or loss to the target object. Another main issue is that most of the analyses in complex sport games are still based on the counting of single events of the game. Those single events are always influenced by the game context and dependent on prior actions. Therefore, during such analyses a lot of information like group tactics gets lost. During the interdisciplinary pilot project "computer supported automated game analysis" at the Institute of Sports Science, Karl-Franzens-University Graz and Institute of Computer Graphics and Vision, Technical University Graz the base has been set to overcome the mentioned problems in game analysis. The aims of the project "Movement and action sequence analysis in complex sport games" are twofold: We develop intelligent software to reduce the hardware demand for multi camera systems which are then used for movement analysis and automated action detection. Furthermore, we combine single events to action sequences to analyze team tactics and to improve success probabilities of action predictions.