Cluster based human action recognition using latent dirichlet allocation

Abstract

Recognizing human actions in video streams is a challenging task in the field of image processing and surveillance. This is due to variabilities in shapes, articulations of human body, cluttered background scene and occlusions. Conventional human action recognition algorithms generate coarse clusters of input videos, with lesser information regarding the cluster generation. In this paper, a mapping technique has been proposed which transforms the gait sequences into document-word template required for topic models such as Latent Dirichlet Algorithm (LDA). LDA is used to group the input videos into finer clusters. Experiments on KTH dataset [10] suggest that the proposed algorithm is effective method for recognizing human actions from the video streams.