Event Detection based on Hidden Conditional Random Field Model in Sport Videos

摘要

This paper proposes a new highlights event detection method for basketball videos. The support feature of each highlight is firstly found using the concept lattice clustering technology according to the audio-video features and middle level semantic features defined in this thesis. Then, the support features are weighted to construct the affective arousal feature. The audio shots are processed to obtain the whistle shots features using the whistle shots detection method defined in this thesis. The affective arousal feature and the whistle shots features are combined as the input. An effective HCRF (Hidden Conditional Random Field) is constructed to realize highlight detection of basketball shooting and fouls. Experimental results show the effectiveness of the proposed method.