Integrating Motion and Shape Analysis for Robust Human Action Recognition Using Spatio-Temporal Dynamics

Abstract

This paper aims at providing an overview of human action recognition with a method that uses the spatio-temporal features extracted from video sequences. Additional features from motion and shape can further improve the action classification and hence the proposed method. Experimental outcomes show that this approach significantly outperforms previous practices for identifying a number of human activities while yielding 100% accuracy on test datasets.