A discriminant approach to sports video classification

The problem of automating sports video classification is investigated by analyzing the low-level visual signal patterns using autocorrelogram. In this paper, two discriminant techniques are tested, namely, neural network with PCA and support vector machine (SVM), when testing data set is larger size than training data set. Seven different kinds of popularly televised sports are studied, namely basketball, Thai boxing, football, golf, diving, tennis, and volleyball. The experiments were emphasized on classifying video sequences at frame level. Classification results indicated that SVM were more efficient supervised learners than neural network with PCA for classifying sports videos with the classification accuracy of up to 91.09%.