SBIR Phase I: Semi-Automated Sports Video Search

点击翻译按钮获取中文摘要

This Small Business Innovation Research Phase I project will develop new technology that will enable precise search of sports videos. Users will be able to search for specific players, teams, and plays from large archives of recorded video sports broadcasts. The proposed research will build on early results of a sports video search engine developed by the team at MIT. The approach combines semantic information mined from speech transcriptions with visual information extracted using video analysis algorithms. The proposed research will extend the existing software algorithms that have been developed for baseball video to other professional and college sports. Additional software tools will be developed to increase the accuracy of the search system, and new user interfaces based on natural language processing algorithms will be designed to enable simplified user access to video. The anticipated result of this research is a method for accurate video search and indexing that enables queries by natural language and requires significantly less human labor to initially tag video than existing techniques. The broader impact of this research comes from the commercialization of this technology as a service layer which provides search and indexing solutions to multiple market segments that together represent a multibillion dollar industry in the United States. The research meets the needs of at least three market segments: (1) Sports professionals, who will gain powerful video access tools enabling better player evaluation, recruiting, coaching, and game analysis; (2) Sports news providers, who will be able to link news stories to related video clips thereby adding value to their media offerings; (3) Sports fans, who will be able to search and browse sports video archives with ease, providing new opportunities for advertising. Initial market research suggests that the access enabled by this technology would have broad impact on how sports video is used. Furthermore, the approach may later be extended to apply beyond sports to other video domains.