Extracting semantic information from basketball video based on audio-visual features

In this paper, we propose a mechanism for extracting semantic information from basketball video sequence using audio and video features. After we divide the input video into shots by a simple cut detection algorithm using visual information, we analyze audio signal data to predict the location of an important event from which a cheering sound happens to start using the combination of MFCC features and the LPC entropy. Finally, we extract semantics about class of shot by computer vision techniques such as basketball tracking and related objects detection. Experimental results show that the proposed scheme can concretely extract semantics from basketball video data as compared to the existing methods.