Improved Video Scene Detection Using Player Detection Methods in Temporally Aggregated TV Sports News

摘要

Many strategies of content-based indexing have been proposed to recognize sports disciplines in sports news videos. It may be achieved by player scenes analyses leading to the detection of playing fields, of superimposed text like player or team names, identification of player faces, detection of lines typical for a given playing field and for a given sports discipline, recognition of player and audience emotions, and also detection of sports objects and clothing specific for a given sports category. The analysis of TV sports news usually starts by the automatic temporal segmentation of videos, recognition, and then classification of player shots and scenes reporting the sports events in different disciplines. Unfortunately, it happens that two (or even more) consecutive shots presenting two different sports events although events of the same discipline are detected as one shot. The strong similarity mainly of colour of playing fields makes it difficult to detect a cut. The paper examines the usefulness of player detection methods for the reduction of undetected cuts in temporally aggregated TV sports news videos leading to better detection of events in sports news. This approach has been tested in the Automatic Video Indexer AVI.