A realtime human action recognition method based on single view key poses in sports video

Abstract

A fast and robust single-view action recognition system is needed in more and more applications, especially on mobile devices. In this paper, we propose a key pose classification based method for action recognition. Firstly, we use a fast feature extraction method to represent these key poses and use a fast boosting method for key pose classification. Secondly, we propose a fast recognition approach based on the results of classification and order constraint. Due to the quick feature extraction method and a fast and robust sequences recognition method, our system can implement real-time pose recognition on mobile devices. Our experiments have indicated that this method has a good performance on many videos in a wild scene.