Change detection in multidimensional data streams with efficient tensor subspace model

Abstract

The paper presents a method for change detection in multidimensional streams of data based on a tensor model constructed from the Higher-Order Singular Value Decomposition of raw data tensors. The method was applied to the problem of video shot detection showing good accuracy and high speed of execution compared with other more time demanding tensor models. In this paper we show two efficient algorithms for tensor model construction and tensor model update from the stream of data.