Audiovisual processing for sports-video summarisation technology

摘要: In this thesis a novel audiovisual feature-based scheme is proposed for the automatic summarization of sports-video content The scope operability designed to encompass wide variety o f sports genres that come under description ‘field-sports’. Given assumption that, in terms conveying narrative field-sports-video, score-update events constitute most significant moments, it their detection should thus yield favourable summarisation solution. To end, generic methodology identification field-sports-video content. based on development robust extractors set critical features, which are shown reliably indicate locations. evidence gathered by feature combined and analysed using Support Vector Machine (SVM), performs event process. An SVM chosen basis its underlying technology represents an implementation latest generation machine learning algorithms, recent advances statistical learning. Effectively, offers solution optimising classification performance decision hypothesis, inferred from given training data. Via phase utilizes 90-hour trainmg-corpus, infers model observing patterns extracted evidence. Using similar but distinct evaluation corpus, effectiveness then tested genencally across multiple fieldsports- video including soccer, rugby, field hockey, hurling, Gaelic football. results suggest task, both high retrieval rejection statistics achievable.