Abstract

Sport has a large number and its number is always increasing as the time goes by. Many people like to watch sports competitions or tournaments on the field live. Nowadays, sports tournaments can now be watched digitally via video without the need to watch them live on the field by technological advances. Video classification needs to be done to differentiate the number of sports videos currently into the sports category according to the content. This study classifies sports videos of the types of Baseball, Basketball, Boxing, Gymnastic, Hockey, Swimming, Tennis, Volleyball, and Wrestling based on image content. The classification is done with the image content as a he, then the superpixel segmentation and superpixel colorization processes are carried out on the images to form a model. The model is formed using the CNN method with ResNet-50 architecture. Various numbers of superpixel segmentation are used as a comparison in the modelling to get the best results. The number of superpixel segmentation with the best result is 25.000 with an average rate of accuracy 0.97, precision 0.87, and recall 0.86 so that it is used for testing sports videos. The test result is the superpixel segmentation can be used to classify sports videos well with an average rate of accuray 0.91, precision 0.64 and recall 0.61, although some sports that had poor results, such as basketball and wrestling.