Sports Video Classification Using Bag of Words Model

摘要

We propose a novel approach classify different sports videos given their groups. First, the SURF descriptors in each key frames are extracted. Then they are used to form the visual word vocabulary (codebook) by using K-Means clustering algorithm. After that, the histogram of these visual words are computed and considered as a feature vector. Finally, we use SVM to train each classifier for each category. The classification result of the video is the production of the scores output from all of the key frames. An extensive experiment is performed on a diverse and challenging dataset of 600 sports video clips downloaded from Youtube with a total of more than 6000 minutes in length for 10 different kinds of sports.