Distributed Data and Information Fusion in Visual Sensor Networks

Computer vision, and in particular multi-camera environments, has been widely researched over the recent years, thus leading to several proposals of multi-camera or visual sensor networks (VSNs) architectures (Valera and Velastin 2005). The aims of these systems are very different; to name some of them, there are examples in surveillance applications (Regazzoni et al. 2001), sport domains (Chen and De Vlesschouwer 2010), or ambient intelligence applications for elderly care (Zhang et al. 2010). Despite the specific goal of each system, all of them have to cope with a distributed architecture of visual sensors to acquire and process information from the environment. The obtained information must then be fused in order to generate a meaningful global picture of the environment. Since a distributed VSN can be applied to different domains/scenarios, a specific ontology provides meaning and sense of the information that the system uses for interpretation purposes. This chapter explores the use of the multi-agent paradigm and ontology-based knowledge representation formalisms to perform distributed data and information fusion (DIF) in VSNs. The multi-agent paradigm, which has been widely applied in distributed systems, provides a theoretical and practical framework to allow communication and cooperation among the components of the system. For instance, in Lesser et al. (2003) several multi-agent protocols are presented to solve the task allocation problem in distributed sensor networks, but without visual capabilities. Classical distributed visual systems work well for monitoring and surveillance tasks, but they can be improved using a multi-agent paradigm and ontology-based mechanisms. The underlying idea is to provide autonomous elements of the system with standard communication capabilities compliant to a content ontology in the process to achieve high-level data fusion