On the Use of MPEG-7 for Visual Surveillance

This paper investigates the use of the MPEG-7 standard to represent the output of automated visual surveillance systems. There are existing standards and proposals for the representation of surveillance outputs in an extensible on tology: the prospects for accommodating their features in a single MPEG-7 surveillance schema are appraised. The aim is to define an extensible framework, suitable for the output of single cameras and whole systems, to be used by operators and automatic processes to satisfy the diverse and extending set of requirements for surveillance metadata. These requirements are discussed in some detail. An MPEG-7 document for a simple surveillance task is constructed, with two layers of descriptors for the observed pedestrians. The first layer comprises colour descriptions of transitory objects observed from single sensors. The sec ond layer aims to provide a uniform, unique identifier for each entity observed in the overall scene. The prototype schema demonstrates techniques to profile and extend the MPEG-7 standard, and uses a graph-based approach to represent the uncertain state of knowledge about the relationships between low and high level elements.