A novel method for online action segmentation and classification

Video based human action has many developments in recent years. Different data sets and algorithms have been evaluated for various approaches. A lot of research work has been published by different authors but still there is need of improvement and amendment. In this article we provide a new algorithm for human body movement which takes in a continuous stream of body poses and performs online action segmentation and classification. Apart from other traditional methods the proposed method does not rely on already segmented test sequences. Obtained results show that the approach performs well on both a per-frame and persegment level. A confidence score is provided for each recognized segment, which can used to isolate previously unseen actions.