The development in automated people tracking technology in the last decade has resulted in applications where large amounts of data can be generated with significantly less manual work than ever before [11, 12, 13]. One of such examples is the computer vision based tracking on sport videos, where, if certain conditions are met, computer tracks athletes with very little user intervention [14]. Such data can be used in performance analysis [10], giving sport community the feedback on player and team performance.

Years ago, analysis of sport matches was almost entirely manual, and the ability to gather certain kinds of data (like motion trajectories) was severely limited [7]. Before the introduction of automatic methods into the sports video analysis, every piece of information had to be entered into the computer by hand [3], and therefore manual annotations generated relatively small amounts of data.

However, by introducing the computer vision based processing of videos, the amount of data may increase dramatically. Widely used video standards (PAL, NTSC) assume frame rates of 25 or 30 video frames per second, and computer vision based tracking methods usually process each captured video frame for the greatest reliability. Therefore, in case of standard PAL videos, motion data of players is available at intervals of 40 milliseconds, and users have the ability to provide their manual annotations at the same temporal resolution, if they wish to. Computer user interfaces with integrated video players provide automatic synchronization of videos to the user annotations and therefore enable users to enter the annotations with much less effort. Additionally, modern desktop personal computers can calculate derived parameters (such as player velocity, acceleration and path length) with neglible computational cost. As a consequence, the use of video based analysis techniques, coupled with computer vision based automatic tracking results in large amounts of output data. The problem is not limited only to video analysis. Alternative means of obtaining player motion data (e.g. radio waves-based localization) may provide even higher position sampling rates.

Download pdf Analysis and pattern detection on large amounts using standard SQL