Abstract: Current research mainly focused on the basics of soccer video, and the work about video semantic analysis was not perfect. We analyzed the interaction behavior in soccer video, and the key problems of soccer video semantic understanding system were studied which made a useful exploration in understanding and analyzing the competition from a common scene. Firstly, the playfield was extracted from the video by using main color of the video and the field line was calibrated. Secondly, the players were detected and judged into the team they belonged to. Simultaneously, ball trajectory tracking was obtained by using Kalman filtering. Lastly, offensive goal of the players was judged, and the real trajectory of the ball was obtained and segmented. Then the offensive strategy was analyzed from the perspective of the soccer field based on the recognition of the interaction behavior between the ball and the players. Experiment results show that the proposed algorithm can effectively solve the problem of each stage in the video analysis and good performance is achieved from analyzing the interactions behavior existed in the video. And the analysis of the offensive strategy is realized from the perspective of the soccer field, the precision of the offense pattern recognition result was more than 71% and the recall rate was above 85%. Good achievements are obtained in both the application of computer vision research and the scientific research for soccer sports.

Key words: offensive strategy, interaction behavior, video semantic understanding,  soccer video