Automatic parsing of sports videos with grammars

Abstract

Motivated by the analogies between languages and sports videos, we introduce a novel approach for video parsing with grammars. It utilizes compiler techniques for integrating both semantic annotation and syntactic analysis to generate a semantic index of events and a table of content for a given sports video. The video sequence is first segmented and annotated by event detection with domain knowledge. A grammar-based parser is then used to identify the structure of the video content. Meanwhile, facilities for error handling are introduced which are particularly useful when the results of automatic parsing need to be adjusted. As a case study, we have developed a system for video parsing in the particular domain of TV diving programs. Experimental results indicate the proposed approach is effective