Sports Classification in Sequential Frames Using CNN and RNN

Abstract

Automatic sports classification is basic but important task for archiving digital contents in broadcasting companies as well as for general video scene understanding. In this paper, deep learning approach combining convolutional and recurrent neural networks is applied to classify five different classes of sports. Features extracted from CNN are combined temporal analysis using RNN. Multiple experiments results show that the effect of different frame sequence correctly classified test data up to 96.66%.