Automatic Event Classification in Baseball Broadcast Videos using Scene Patternization Focusing on Post-Pitch Shot and Discrete Hidden Markov Models

Abstract

A method has been developed for automatically classifying baseball video scenes into some events that describe their content.The baseball scenes are patternized using a set of rectangles with image features and motion vectors.The basic unit for patternization is a shot.For the second shot of each scene which includes significant information for event-classification,a partial shot generated by dividing the shot is used as a processing unit.The scenes used for training are expressed as sequenced symbols based on the patternized data for shots and partial shots.“Event-unknown”baseball scenes are assigned “event-indexes”(i.e.,homerun,single,walk,etc.) using discrete hidden Markov models that have been trained with the training symbol sequences for each kind of event.An experiment using videos of seven Major League Baseball games produced good results,demonstrating that this method can automatically classify events with high accuracy.