Workload modeling for multimedia surveillance systems

Abstract

Amultimedia surveillance system has to sustain high computational loads. Such a system needs to intelligently learn the characteristics of the workload that it is going to handle. The knowledge of the workload also provides a strong basis for design and optimization of the system components. To efficiently use this knowledge, we need an analytical model of the workload. The traditional multimedia workload models used in other domains are not appropriate for surveillance systems. In other domains, the workload characteristics are mainly derived from the statistical properties of the data,whereas in the case of surveillance, the semantics play a dominant role in determining the processing needs. In this chapter, we discuss popular workload models from other domains and explore their applicability to surveillance systems. We find that none of those models describe the workload accurately in surveillance context. Following this observation, we propose a novel Markov chain based formal model of multimedia workload for surveillance systems. Different states of the Markov chainmeticulously capture the variability of the workload. Themodel is validated with real surveillance data. Subsequently, we describe performance analysis of a real surveillance system based on the proposed model. © Springer-Verlag Berlin Heidelberg 2013.