Detailed analyses of game situations are necessary for the planning and controlling of training and the development of successful game tactics. The aim of the present study is to minimize manual effort in the analysis of sprint and jumping movements. Automated action recognition is exemplarily applied in beach volleyball. Modern video analysis methods are used to generate position data from which sprints and jumps can be detected automatically. Methods: To test the determination of sprints and jumps from professional beach volleyball athletes, action paths from 22 rallies at the World Tour Tournament 2007 in Klagenfurt were analyzed. By using a tracker based on particle filters the position of the athlete`s upper bodies was calculated from video. A further calculation step allowed determining the lower body segments to get the athlete`s position at the court (Mauthner & Bischof, 2007; Mauthner et al., 2007). Jumping movements could be detected automatically from calculated movement velocities and their variations in time. Errors in the position data due to jumping movements were automatically corrected by linear interpolation. Sprints (defined as movement velocities > 4.16 m/s) were then detected from the corrected data. Results: Preliminary results show, that the players execute 0.75 ± 0.64 jumps and 0.27 ± 0.45 sprints per rally. These automated detected jump actions exceed the manual counted by 2.6%, whereas 5.9% false positives and 8.8% false negatives occurred. The detected sprint actions exceed the manual data by 10% derived from 23.1% false positives and 38.5% false negatives. Discussion: The tests of the described method show the principal possible application of video based techniques for automatic detection of movements like sprints and jumps. However, the amount of tested data still lacks of representative quantity to describe the game structure in beach volleyball. Differences to manually determined values in jump detection are due to inaccuracy of the tracking method, especially from game scenes which are occluded by the net. In the future, improved tracking algorithms combined with adequate filters may significantly improve results. The substantial differences in sprint detection were mainly due to subjective manual assessment of sprint action. The automated sprint detection may objectify results.
© Copyright 2009 14th annual Congress of the European College of Sport Science, Oslo/Norway, June 24-27, 2009, Book of Abstracts. Veröffentlicht von The Norwegian School of Sport Sciences. Alle Rechte vorbehalten.