Semantic Analysis in Sports Video

Abstract

Sports video has been widely studied due to its tremendous commercial potentials. Therefore this work designs a high-level annotation system for tennis video to profit contents retrieval. First frames that contain tennis court-scenes are extracted and then the 13 important points on the court, defined by the proposed tennis court-point model, are detected and tracked, along with the locations of the two tennis players. The relative positions of the two players on the tennis court are analyzed using a high-level reasoning model to annotate video clips. Finally, the court of the image is mapped to the court model, using perspective projection to recover the shape of the tennis court. The results in VRML are shown in the experimental section.