Portal de Periódicos da CAPES

False and Miss Detections in Temporal Segmentation of TV Sports News Videos – Causes and Remedies

2014; Springer Nature; Linguagem: Inglês

10.1007/978-3-319-10383-9_4

2194-5357

Kazimierz Choroś,

Music and Audio Processing

Categorization of sports videos, i.e. the automatic recognition of sports disciplines in videos, mainly in TV sports news, is one of the most important process in content-based video indexing. It may be achieved using different strategies such as player scenes analyses leading to the detection of playing fields, recognition 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, detection of sports objects specific for a given sports category, and also recognition of player and audience emotions. The sports video indexing usually starts with the automatic temporal segmentation. Unfortunately, it could happen that two or even more consecutive shots of two different sports events, although most frequently of the same sport discipline, are falsely identified as one shot. The strong similarity mainly of color of playing fields does not sometimes allow the detection of a video transition. On the other hand, very short shots of several frames are detected in case of dissolve effects or they are simply false detections. Most often it is due to very dynamic movements of players or a camera during the game, changes in advertising banners on playing fields, as well as due to light flashes. False as well as miss detections diminish the efficacy of the temporal aggregation method applied to video indexing. The chapter examines the cases of false and miss detections in temporal segmentation of TV sports news videos observed in the Automatic Video Indexer AVI. The causes and remedies of false and miss detections are discussed.