Indexing of baseball telecast for content-based video retrieval

We propose a method to index baseball telecast for content-based video retrieval. In our approach, assumptions specific to baseball telecast are used to index video-recordings. In, the first stage of the system, we detect domain specific scene in a baseball video based-on image similarity. These scene, called basic scene in the paper, are the shots which include a single pitching in each. After extracting these scenes, we spot the exact location of pitching and batting action using continuous dynamic programming matching for fixed areas in the image. If the batter swing the bat, we determine the end point of the play from the camera view after batting to recognize the batting result. We also recognize the caption to verify/confirm the recognition result. We stored summarized version of the telecast with these indexes to form a video-database. To validate the approach, we parsed 53 throws in a baseball game using the system, and the percentages of correct spotting for pitching and batting are 96% and 89%, respectively. The validity of the indexing is examined in the report.