Sports Video Classification Using Convolutional Neural Network (CNN) with Normalization Flow
Nurul A. Emran, Nurul Izrin Md Saleh, Muhamad Zaidi Mohd Ali · 2024
Classifying sports videos is a crucial task in various applications, especially sports analytics, video retrieval, and content analysis. Unlike image classification, video classification is more challenging due to the dynamic nature of videos. Advancements in deep learning, particularly Convolutional Neural Network (CNN), have shown promising results in sports video classification. In this paper, we present the results of implementing CNN with normalization flow to classify several sports classes. In particular, the effect of classification accuracy in training datasets by increasing the number of classes (with similar characteristics and some noise) is analyzed. The effect of frame averaging, and the number of epochs were also observed. The results show that for the training dataset, CNN performance is slightly affected by the additional classes, but by increasing the number of epochs, the accuracy of training and validation datasets has improved. CNN can still maintain highly accurate classifications in test datasets (more than 80%) in some observations. CNN with frame averaging shows lower errors than single-frame CNN.