Semantic units detection and summarization of baseball videos

Abstract

A framework for analyzing baseball videos and generation of game summary is proposed. Due to the well-defined rules of baseball games, the system efficiently detects semantic units by the domain-related knowledge, and therefore, automatically discovers the structure of a baseball game. After extracting the information changes that are caused by some semantic events on the superimposed caption, a rule-based decision tree is applied to detect meaningful events. Only three types of information, including number-of-outs, score, and base occupation status, are taken in the detection process, and thus the framework detects events and produces summarization in an efficient and effective manner. The experimental results show the effectiveness of this framework and some research opportunities about generating semantic-level summary for sports videos.