Vision based semantic analysis of surveillance videos

Abstract

As recent research in automatic surveillance systems has attracted many cross-domain researchers, a large-number of algorithms have been proposed for automating surveillance systems. The objective of this chapter is twofold: First, we present an extensive survey of different techniques that have been proposed for surveillance systems categorised into motion analysis, visual feature extraction and indexing. Second, an integrated surveillance framework for unsupervised object indexing is developed to study and evaluate the performance of visual features. The study focuses on two characteristics highly related with human visual perception, colour and texture. The set of visual features under analysis comprises two categories, new leading visual features versus state-of-the-art MPEG-7 visual features. The evaluation of the framework is carried out with AVSS 2007 and CamVid 2008 datasets. © 2013 Springer-Verlag Berlin Heidelberg.