Collaborative target classification with multiagent system in wireless multimedia sensor networks

Recent technological advances in hardware design and wireless communications, together with the availability of low cost microphones and video camera, have stimulated the emergence of wireless multimedia sensor networks (WMSNs). This paper proposes to explore such novel approaches as inspired by multi-agent system (MAS) researches that conduct systematic investigation of interaction between autonomous entities. Within the MAS framework, a hierarchical WMSN architecture is established and swarm intelligence is introduced to facilitate its formation. Contract net protocol is employed to achieve efficient resource allocation among the sensor nodes for the purpose of intruding target classification. To make the best global decision, the mechanism of committee voting is exploited to combine all the individual decisions. The supervised machine learning techniques of Gaussian process classifier is presented for investigating the performance of the proposed hierarchical WMSN architecture. Simulation experiments with real-world data have been extensively performed to evaluate the proposed architecture, resource allocation and decision fusion mechanisms, and the results show that the proposed MAS approaches are effective and efficient.