Basketball game systems identification by real time video processing

In this paper, we describe a new strategy of observation and analysis of a basketball match, exploiting video processing techniques in order to identify the game system of a team. Our system realizes the players′ positions tracking during the match. We have in entry three video streams from three fixed cameras, each processed separately to deliver measures of players′ positions from different available views. Each treated view includes foreground detection and a bounding box tracker aimed at containing pixels standing for the players. During the multi-view process, measures resulting from different sights are synchronized in order to identify the same player who can be seen simultaneously on many cameras. These measures are combined in order to obtain the players′ positions as well as their updates through the images. The positions thus obtained have been exploited in a database containing the representative points (coordinates) of all the players who form a polygon. The analysis of a game system is thus nothing but the analysis of the distortion and movement of this polygon during the match. We define comparative indicators of the two teams as well as an indicator which represents a discipline expert′s opinion (action code) and we exploit statistical tools whose objectives are on the one hand to emit correlations and relationships between different indicators, and on the other hand to identify the game system adopted by comparing the expert opinion with the results of the established heuristic modals.