Sensor Relevance Establishment Problem in SharedInformation Gathering Sensor Networks

In a city area network hundreds of video cameras, infrared and laser sensors are deployed for online monitoring of physical phenomenon over a geographical area. This is a popular application of sensor networks. Next generation intelligent sensing systems and networks are divided into two categories, an always-on mode -where every sensor information is piped to a base station (for resolution of a problem), and a snapshot mode -where a user queries the network for an instantaneous summary of the observed environment. Snapshot mode sensor networks are highly dependent on relevant sensing due to the accuracy required in a short time and the sensitive nature of the problem (query). This paper summarises the sensor relevance establishment problem in data acquisition. We describe its use in a framework that models the observed environment at each sensor node as a function of time, and uses an adaptive learning method to sample data with the corresponding relevance metric. We take the sensor network towards the problem by considering the relevance metric at given time step. The sensor relevance establishment problem has been split into two steps. In step one, the relevant sensor type is discovered based upon the IEEE 1451.4 Transducers Electronic Data Sheets (TEDS). TEDS description model can be used to discover the sensor type and their geographical locations and other important information such as uncertainty measurement functions and information fusion rules necessary to fuse multi-sensor data. In step two, the most useful sensor selection is determined using the relevant information data metric. This step is modelled using the Kullback-Leibler Divergence (KLD) method to measure the information relevance distance between the TEDS modelled relevant sensors determined in step one. As proof of our concept we have simulated the 3D environment using a real-time distributed robotics software Player/Stage/Gazebo. The preliminary results have been demonstrated on a simple autono- mous robot navigation problem.