Representative Based Clustering of Long Multivariate Sequences with Different Lengths

Abstract

Video streams as unstructured or poorly structured data issue a challenge to create a unified framework capable to depict and convey high-level stories. Up-to-date indexing and search techniques to manage video data are able to operate the voluminous amounts of contained in video information in order to detect spatial and temporal events. Nevertheless, bridging semantic gap between the low-level frame or video features and high-level semantic concepts necessitates extremely high-speed procedures of temporal unlabeled data. Automatic video annotation in visual forms appears one of the promising approaches representing most pertinent and crucially important information. This goal is achieved by (among others) clustering large collections of video data.