A grammar based method for video event indexing and retrieval

Abstract

In this paper, we present a novel framework for indexing and retrieving video events. The framework has a hierarchical structural and based upon a grammatical model. We firstly define and detect some event primitives, and induce the parameters of the SCFG to describe the event sequences, by using Liang's nonparametric model of HDP-SCFG and variational inference algorithm. The MES parser and viterbi algorithm is employed to acquire the parse tree, which is index and matched next. We mainly contribute in four aspects. 1) It is the first time that transplant nonparametric grammar, ISCFG, to the domain of video event retrieval. 2) The cross data link table structure of indexing is efficient enough for searching in real time. The computational complexity of locating a non-terminal at any time point is O(log). 3) The novel matching scheme allows people to search content of interest by input a sample of video segment. The measure of similarity works well because both the structure and content of the parse trees are took into account. © 2012 Springer-Verlag GmbH.