Short track speed skating has become popular all over the world. The demands of a computer-aided training (CAT) system are booming due to this fact. However, the existing commercial systems for sports are highly dependent on expensive equipment and complicated hardware calibration. This dissertation presents a novel CAT system for tracking multiple skaters in short track skating competitions. Aiming at the challenges, we utilize global rink information to compensate camera motion and obtain the global spatial information of skaters; apply Random Forest to fuse multiple cues and predict the blobs for each of the skaters; and finally develop a silhouette and edge-based template matching and blob growing method to allocate each blob to corresponding skaters. The proposed multiple skaters tracking algorithm organically integrates multi-cue fusion, dynamic appearance modeling, machine learning, etc. to form an efficient and robust CAT system. The effectiveness and robustness of the proposed method are presented through experiments.
© Copyright 2014 Published by Utah State University. All rights reserved.