Recognition of human actions from RGB-D videos using a reject option

Abstract

In this paper we propose a method for recognizing human actions by using depth images acquired through a Kinect sensor. The depth images are represented through the combination of three sets of well-known features, respectively based on Hu moments, depth variations and the transform, an enhanced version of the Radon transform. A GMM classifier is adopted and finally a reject option is introduced in order to improve the overall reliability of the system. The proposed approach has been tested over two datasets, the Mivia and the MHAD, showing very promising results. © 2013 Springer-Verlag.