Abstract: The Semantic Gap which exists between video low-level features and high-level semantics is the bottleneck of the video analysis and retrieval. To bridge the gap in football matches videos, the graphical model was brought in, and the Video Semantic Graph was proposed to modeling the high-level semantics. In addition, a DFS-based VSG matching algorithm was proposed, the time complexity and the space complexity were analyzed. The results of the experiment demonstrate the validity of the presented method，which can solve the problems of video semantic modeling and matching.

Key words: Video semantic graph, Graphical model, Graph model matching