A noval method on action recognition based on region and speed of skeletal end points

This paper proposes a method to recognize human actions from a video sequence. The actions include walking, running, jogging, hand waving, clapping and boxing. The actions are categorized after recognition using a decision tree. Apart from other algorithms, our proposed method recognizes single human actions considering the speed, direction and the percentage of endpoints as a novel approach. In addition to action recognition this paper also proposes an error correction method for removing bifurcation. The system has been checked on various datasets and it performed well.