Computer vision-based systems have been applied to sports broadcasting and sports analysis. In particular, object detection and tracking algorithms have been extensively adopted because they can be used to automatically annotate game elements. Curling is a highly strategic sport. The strategic is determined by a curling stone`s trajectory and position. This paper presents the automatic annotation of curling stone position in broadcast videos, for which a mean-shift tracking algorithm is used to examine the appearance and colour information of curling stones. The algorithm uses a multi-dimensional histogram feature vector that contains a sampled pixel value of each RGB colour channel and edge information. The mean-shift kernel is a circular kernel. We also use a Kalmanfilter to track occluded stones. The experimental results show that the proposed method more precisely and rapidly tracks stones than does a general mean-shift method.
© Copyright 2015 Proceedings of the World Congress on Electrical Engineering and Computer Systems and Science (EECSS 2015). Published by Avestia Publishing. All rights reserved.