Comparison of Clustering Methods for Pose Based Video Summarization

The aim of this paper is to compare and evaluate different methods for clustering human action poses for video summarization. In this respect, three different clustering approaches are compared. These are the commonly known clustering algorithm "K-means", a spectral clustering method "Normalized Cuts" and a new clustering method "Affinity Propagation". These algorithms are utilized and compared with respect to their performance on clustering action poses on videos that contain different human actions. The experimental results demonstrate that k-means algorithm is more effective for the purpose of pose clustering and video summary generation.