Silhouette based human action recognition using PCA and ISOMAP

Abstract: Human action recognition in computer vision has attracted strong research interest in recent years because of its promising applications. Automated detection of a person‘s atomic movements is called human action recognition. Diverse methods have been proposed for detecting the human action due to the dynamics in the underlying representation of human body. This paper provides a glance over the significant researches conducted in this area and a comprehensive survey of the different taxonomies suggested by different researchers. A system is proposed for efficient action recognition based on the concept ‗bag of correlated poses‘ by taking into account of the correlation between sequential poses in an action on both linear and non linear feature vectors.

Keywords: Action recognition, Correlogram of body poses, PCA, ICA, ISOMAP.