Automatic composition of broadcast sports video

摘要: This study examines an automatic broadcast soccer video composition system. The research is important as the ability to automatically compose sports will not only improve generation efficiency, but also provides possibility customize broadcasting. We present a novel approach two major issues required in system's implementation, specifically camera view selection/switching module and replay module. In our we use multi-modal framework perform content analysis, event boundary detection from raw unedited main/sub-camera captures. explores possible cues using mid-level representations bridge gap between low-level features high-level semantics. analysis results are utilized for generated composition, used generate replays which inserted into video. Our experimental promising found be comparable those by professionals.