Soccer videos analysis based on the active discriminant model

Since the beginning of the 21th century, with the continuous development of information technology, it has been the difficulty and hotpot in understanding field to study and analyze sports videos with computer technology. Because soccer videos have a broad mass background, it is more valuable to research on it. At present, there are many researches based on trajectory information of players or balls. But due to the background interference and other reasons, the detection accuracy still needs to be improved. In this paper, we present a method of tactical behavior recognition based on the local spatio-temporal regression kernel according to the activity of the players who execute tactics on the playfield. Firstly, we detected the playfield and divided it into some parts. Secondly, we established an active discrimination model by using the local spatio-temporal regression kernel as a future detector for the detection and location of active players, and constructing future bag models. Finally, we divided the tactical behaviors into six classes to realize automatic recognition of them. The method of this paper is used to the video segments of the associated goal events of the 2012 European Cup and the 2013/14 season of Barcelona in the La Liga, and the average accuracy rate of the experiments is 91.3%.